feat(route): add Meta AI global search#21839
Open
shcheglovnd wants to merge 3 commits intoDIYgod:masterfrom
Open
feat(route): add Meta AI global search#21839shcheglovnd wants to merge 3 commits intoDIYgod:masterfrom
shcheglovnd wants to merge 3 commits intoDIYgod:masterfrom
Conversation
Adds /meta/ai/global-search/:routeParams? backed by the useFBAIGlobalSearchQuery GraphQL query (doc_id 9716930201759979). Filters supported via path-encoded routeParams: q, content_types (person/publication/blog/dataset/event/tool), research_areas, filter_tags, years, location_cities, alphabetical_filter, sort_by (RELEVANCE/MOST_RECENT/ALPHABETICAL/RANDOM), offset. Limit stays as ?limit=N because the cache layer keys on it. Path-based filters keep each combination in its own cache entry, matching the routeParams pattern used by the weibo/keyword route. Extracts the LSD/SiteData fetch and GraphQL body builder into routes/meta/utils.ts so ai-blog.ts and ai-global-search.ts share the boilerplate.
github-actions
Bot
added
route
auto: route no found
Contributor
|
Please use actual values in |
CodeFactor flagged the handler as a Complex Method (cyclomatic complexity 16). Extract buildSearchInput, summarizeFilters, and mapItem helpers so handler becomes a thin orchestrator and each helper has a small, isolated branch count. No behavior change; verified locally against the same filter combinations as the original (default, single content_type, multi-filter with q+years+sort_by, alphabetical+person).
github-actions
Bot
removed
the
auto: route no found
Contributor
Auto Review
|
Contributor
|
Successfully generated as following: http://localhost:1200/meta/ai/global-search - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/ (sort: RELEVANCE, total hits: 2848). - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:09:43 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>AIRA₂: Overcoming Bottlenecks in AI Research Agents</title>
<description>Existing research has identified three structural performance bottlenecks in AI research agents: (1) synchronous single-GPU execution constrains sample throughput, limiting the benefit of search; (2) a generalization gap where validation-based selection causes overfitting and performance to degrade over extended search horizons; and (3) the limited capability of fixed, single-turn LLM operators imposes a ceiling on search performance. We introduce AIRA₂, which addresses these bottlenecks through three architectural choices: an asynchronous multi-GPU worker pool that increases experiment throughput linearly; a Hidden Consistent Evaluation protocol that delivers a reliable evaluation signal; and ReAct agents that dynamically scope their actions and debug interactively. On MLE-bench-30, AIRA₂ achieves a mean Percentile Rank of 81.5% at 24 hours and 83.1% at 72 hours, outperforming the strongest baseline, which achieves 72.7%. On AIRS-Bench, AIRA₂ exceeds human state-of-the-art on 6 out of 20 diverse research tasks. Ablations confirm that each architectural component is necessary, that performance follows a predictable scaling law that transfers across LLM backbones, and that the "overfitting" reported in prior work was driven by evaluation noise rather than true data memorization.</description>
<link>https://ai.meta.com/research/publications/aira-overcoming-bottlenecks-in-ai-research-agents/</link>
<guid isPermaLink="false">1298625375513718</guid>
<pubDate>Thu, 16 Apr 2026 07:00:00 GMT</pubDate>
<author>Karen Hambardzumyan, Nicolas Baldwin, Edan Toledo, Rishi Hazra, Michael Kuchnik, Bassel Al Omari, Thomas Simon Foster, Anton Protopopov, Jean-Christophe Gagnon-Audet, Ishita Mediratta, Kelvin Niu, Michael Shvartsman, Alisia Lupidi, Alexis Audran-Reiss, Parth Pathak, Tatiana Shavrina, Despoina Magka, Hela Momand, Derek Dunfield, Nicola Cancedda, Pontus Stenetorp, Carole-Jean Wu, Jakob Foerster, Yoram Bachrach, Martin Josifoski</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
</item>
<item>
<title>TransText: Transparency Aware Image-to-Video Typography Animation</title>
<description>We introduce the first method, to the best of our knowledge, for adapting image-to-video models to layer-aware text (glyph) animation, a capability critical for practical dynamic visual design. Existing approaches predominantly handle the transparency-encoding (alpha channel) as an extra latent dimension appended to the RGB space, necessitating the reconstruction of the underlying RGB-centric variational autoencoder (VAE). However, given the scarcity of high-quality transparent glyph data, retraining the VAE is computationally expensive and may erode the robust semantic priors learned from massive RGB corpora, potentially leading to latent pattern mixing. To mitigate these limitations, we propose TransText, a framework based on a novel Alpha-as-RGB paradigm to jointly model appearance and transparency without modifying the pre-trained generative manifold. TransText embeds the alpha channel as an RGB-compatible visual signal through latent spatial concatenation, explicitly ensuring strict cross-modal (RGB-and-Alpha) consistency while preventing feature entanglement. Our experiments demonstrate that TransText significantly outperforms baselines, generating coherent, high-fidelity transparent animations with diverse, fine-grained effects.</description>
<link>https://ai.meta.com/research/publications/transtext-transparency-aware-image-to-video-typography-animation/</link>
<guid isPermaLink="false">2172593836888174</guid>
<pubDate>Tue, 14 Apr 2026 07:00:00 GMT</pubDate>
<author>Fei Zhang, Zijian Zhou, Bohao Tang, Sen He, Hang Li (BizAI), Zhe Wang, Soubhik Sanyal, Pengfei Liu, Viktar Atliha, Tao Xiang, Frost Xu, Semih Gunel</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Computer Vision</category>
<category>ML Applications</category>
</item>
<item>
<title>Think in Strokes, Not Pixels: Process-Driven Image Generation via Interleaved Reasoning</title>
<description>Humans paint images incrementally: they plan a global layout, sketch a coarse draft, inspect, and refine details, and most importantly, each step is grounded in the evolving visual states. However, can unified multimodal models trained on text-image interleaved datasets also imagine the chain of intermediate states? In this paper, we introduce process-driven image generation, a multi-step paradigm that decomposes synthesis into an interleaved reasoning trajectory of thoughts and actions. Rather than generating images in a single step, our approach unfolds across multiple iterations, each consisting of 4 stages: textual planning, visual drafting, textual reflection, and visual refinement. The textual reasoning explicitly conditions how the visual state should evolve, while the generated visual intermediate in turn constrains and grounds the next round of textual reasoning. A core challenge of process-driven generation stems from the ambiguity of intermediate states: how can models evaluate each partially-complete image? We address this through dense, step-wise supervision that maintains two complementary constraints: for the visual intermediate states, we enforce the spatial and semantic consistency; for the textual intermediate states, we preserve the prior visual knowledge while enabling the model to identify and correct prompt-violating elements. This makes the generation process explicit, interpretable, and directly supervisable. To validate proposed method, we conduct experiments under various text-to-image generation benchmarks.</description>
<link>https://ai.meta.com/research/publications/think-in-strokes-not-pixels-process-driven-image-generation-via-interleaved-reasoning/</link>
<guid isPermaLink="false">847030005080644</guid>
<pubDate>Thu, 09 Apr 2026 07:00:00 GMT</pubDate>
<author>Lei Zhang, Junjiao Tian, Zhipeng Fan, Kunpeng Li, Jialiang Wang, Weifeng Chen, Markos Georgopoulos, Felix Xu, Yuxiao Bao, Julian McAuley, Manling Li, Zecheng He</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Computer Vision</category>
</item>
<item>
<title>Scaling How We Build and Test Our Most Advanced AI</title>
<description>As we build more capable, personalized AI, reliability, security, and user protections are more important than ever. </description>
<link>https://ai.meta.com/blog/scaling-how-we-build-test-advanced-ai/</link>
<guid isPermaLink="false">4369469056623958</guid>
<pubDate>Wed, 08 Apr 2026 17:00:17 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/668130848_1453339552354697_3291493920555502062_n.png?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=yvFttOBoRYkQ7kNvwE2zAxR&_nc_oc=Adr2UifKLIVIje1oDq_d_Zzzd2xhHfN26iDMLvatYMqMAKMUrJtd8QiHDdByuQsgxo8&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0BM5OU8FJ5DWsya0SAnnxVY-9oLCNuVzCJniDz25spnA&oe=6A08CDEE" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Introducing Muse Spark: Scaling Towards Personal Superintelligence</title>
<description></description>
<link>https://ai.meta.com/blog/introducing-muse-spark-msl/</link>
<guid isPermaLink="false">963191949487556</guid>
<pubDate>Wed, 08 Apr 2026 16:00:59 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/667507363_1871749950191919_4996140970371485757_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=IbntS1cBAcsQ7kNvwFGpZpf&_nc_oc=AdoWRIZVlnRJCHKphpg-44LJdWpIcGyseuyTtGVMwVPvYn8IC_2M7OqrJOG1JaQJSk0&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af3t-0wQELUoZHdXQzI6YLBhSnIRsAaeWIRZ2liU1I5cvA&oe=6A08D22C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>How Alta Daily Uses Meta’s Segment Anything to Reimagine the Digital Closet</title>
<description></description>
<link>https://ai.meta.com/blog/alta-daily-fashion-app-segment-anything/</link>
<guid isPermaLink="false">2194788291268065</guid>
<pubDate>Mon, 06 Apr 2026 18:53:27 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/658967078_993415586351079_325919381939729048_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=cBKF2Wk0XYoQ7kNvwHg8n78&_nc_oc=Adpp8hWFe9GVSkrNsSGgcTK4r7FP6jRCnFsR4SVl4s8_NG3Td8pcFrQWplQHD9AJ0_Y&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af1r-86WulkS_07y-FBKw65rpltRtMTr0ex3arrOZ9gLzw&oe=6A08B252" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning</title>
<description></description>
<link>https://ai.meta.com/blog/segment-anything-model-3/</link>
<guid isPermaLink="false">715047027731045</guid>
<pubDate>Fri, 27 Mar 2026 16:00:37 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/583339165_852765810541557_5592130597432767690_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=WvtkkpGzh44Q7kNvwEubR7-&_nc_oc=AdosYwdove82TFwfnCLk3q5bXi6OYKYKFlbayU_okqqiX8vtIP70WBbgYw_OqvZ58J8&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0H-eTtR6ySYweEo716GXNYWl5VptW8MKhFKfPyMVhORQ&oe=6A08D061" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>Introducing TRIBE v2: A Predictive Foundation Model Trained to Understand How the Human Brain Processes Complex Stimuli</title>
<description></description>
<link>https://ai.meta.com/blog/tribe-v2-brain-predictive-foundation-model/</link>
<guid isPermaLink="false">25512119895129580</guid>
<pubDate>Thu, 26 Mar 2026 16:00:16 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/659112495_952140057346056_1728296311373871062_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=rBe_VZxcFD4Q7kNvwGOhdP7&_nc_oc=AdoaYP1BwQzfnCGpSQaVG8V8CaMbbznHgw9z7Mt3CIaeiUr0iUJpZ_p5I6em3s7sWpk&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af2ZpaRD6ER3bB3b0VbZgwp0D5rd_ubqpnajgcv7zhl7SQ&oe=6A08AF6B" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Research</category>
</item>
<item>
<title>A foundation model of vision, audition, and language for in-silico neuroscience</title>
<description>Cognitive neuroscience is fragmented into specialized models, each tailored to specific experimental paradigms, hence preventing a unified model of cognition in the human brain. Here, we introduce TRIBE v2, a tri-modal (video, audio and language) foundation model capable of predicting human brain activity in a variety of naturalistic and experimental conditions. Leveraging a unified dataset of over 1,000 hours of fMRI across 720 subjects, we demonstrate that our model accurately predicts high-resolution brain responses for novel stimuli, tasks and subjects, superseding traditional linear encoding models, delivering several-fold improvements in accuracy. Critically, TRIBE v2 enables in silico experimentation: tested on seminal visual and neuro-linguistic paradigms, it recovers a variety of results established by decades of empirical research. Finally, by extracting interpretable latent features, TRIBE v2 reveals the fine-grained topography of multisensory integration. These results establish artificial intelligence as a unifying framework for exploring the functional organization of the human brain.</description>
<link>https://ai.meta.com/research/publications/a-foundation-model-of-vision-audition-and-language-for-in-silico-neuroscience/</link>
<guid isPermaLink="false">1009839085553251</guid>
<pubDate>Thu, 26 Mar 2026 07:00:00 GMT</pubDate>
<author>Stéphane d'Ascoli, Jérémy Rapin, Yohann Benchetrit, Teon Brooks, Katelyn Begany, Josephine Raugel, Hubert Jacob Banville, Jean Remi King</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
</item>
<item>
<title>HyperAgents</title>
<description>Self-improving AI systems aim to reduce reliance on human engineering by learning to improve their own learning and problem-solving processes. Existing approaches to recursive self-improvement typically rely on fixed, handcrafted meta-level mechanisms, which fundamentally limit how fast such systems can improve. The Darwin Gödel Machine (DGM)(Zhang et al., 2025b) demonstrates that open-ended self-improvement is achievable in coding. Starting from a single coding agent, the DGM repeatedly generates and evaluates self-modified variants, forming a growing archive of stepping stones for future improvement. Because both evaluation and self-modification are coding tasks, gains in coding ability can translate into gains in self-improvement ability. However, this alignment does not generally hold beyond coding domains.
We introduce hyperagents, self-referential agents that integrate a task agent (which solves the target task) and a meta agent (which modifies itself and the task agent) into a single editable program. Crucially, the meta-level modification procedure is itself editable, enabling metacognitive self-modification, improving not only task-solving behavior, but also the mechanism that generates future improvements. We instantiate this framework by extending DGM to create DGM-Hyperagents (DGM-H). By allowing the improvement procedure to evolve, the DGM-H eliminates the assumption of domain-specific alignment between task performance and self-modification skill, and can potentially support self-accelerating progress on any computable task. Across diverse domains (coding, paper review, robotics reward design, and Olympiad-level math-solution grading), the DGM-H improves performance over time and outperforms baselines without self-improvement or open-ended exploration, as well as prior self-improving systems like DGM. We further show that the DGM-H improves the process by which it generates new agents (e.g., persistent memory, performance tracking), and that these meta-level improvements transfer across domains and accumulate across runs. All experiments were conducted with safety precautions (e.g., sandboxing, human oversight). We discuss what safety entails in this setting and the broader implications of self-improving systems. DGM-Hyperagents offer a glimpse of open-ended AI systems that do not merely search for better solutions, but continually improve their search for how to improve.</description>
<link>https://ai.meta.com/research/publications/hyperagents/</link>
<guid isPermaLink="false">1669124537439877</guid>
<pubDate>Tue, 24 Mar 2026 07:00:00 GMT</pubDate>
<author>Jenny Zhang, Bingchen Zhao, Winnie Yang, Jakob Foerster, Sam Devlin, Tatiana Shavrina</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
<category>Open Source</category>
</item>
<item>
<title>Omnilingual MT: Machine Translation for 1,600 Languages</title>
<description>Advances made through No Language Left Behind (NLLB) have demonstrated that high-quality machine translation (MT) scale to 200 languages. Later Large Language Models (LLMs) have been adopted for MT, increasing in quality but not necessarily extending language coverage. Current systems remain constrained by limited coverage and a persistent generation bottleneck: while crosslingual transfer enables models to somehow understand many undersupported languages, they often cannot generate them reliably, leaving most of the world’s 7,000 languages—especially endangered and marginalized ones—outside the reach of modern MT. Early explorations in extreme scaling offered promising proofs of concept but did not yield sustained solutions.
We present Omnilingual Machine Translation (OMT), the first MT system supporting more than 1,600 languages. This scale is enabled by a comprehensive data strategy that integrates large public multilingual corpora with newly created datasets, including manually curated MeDLEY bitext, synthetic backtranslation, and mining, substantially expanding coverage across long-tail languages, domains, and registers. To ensure both reliable and expansive evaluation, we combined standard metrics with a suite of evaluation artifacts: BLASER 3 quality estimation model (reference-free), OmniTOX toxicity classifier, BOUQuET dataset (a newly created, largest-to-date multilingual evaluation collection built from scratch and manually extended across a wide range of linguistic families), and Met-BOUQuET dataset (faithful multilingual quality estimation at scale). We explore two ways of specializing an LLM for machine translation: as a decoder-only model (OMT-LLaMA) or as a module in an encoder–decoder architecture (OMT-NLLB). The former consists of a model built on LLaMA3, with multilingual continual pretraining and retrieval-augmented translation for inference-time adaptation. The latter is a model built on top of a multilingual aligned space (OmniSONAR, itself also based on LLaMA3), and introduces a training methodology that can exploit non-parallel data, allowing us to incorporate the decoder-only continuous pretraining data into the training of an encoder–decoder architecture. Notably, all our 1B to 8B parameter models match or exceed the MT performance of a 70B LLM baseline, revealing a clear specialization advantage and enabling strong translation quality in low-compute settings. Moreover, our evaluation of English-to-1,600 translations further shows that while baseline models can interpret undersupported languages, they frequently fail to generate them with meaningful fidelity; OMT-LLaMA models substantially expand the set of languages for which coherent generation is feasible. Additionally, OMT models improve in cross-lingual transfer, being close to solving the “understanding” part of the puzzle in MT for the 1,600 evaluated. Beyond strong out-of-the-box performance, we find that finetuning and retrieval-augmented generation offer additional pathways to improve quality for the given subset of languages when targeted data or domain knowledge is available. Our leaderboard and main humanly created evaluation datasets (BOUQuET and Met-BOUQuET) are dynamically evolving towards Omnilinguality and freely available.</description>
<link>https://ai.meta.com/research/publications/omnilingual-mt-machine-translation-for-1600-languages/</link>
<guid isPermaLink="false">1449924233245450</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual MT Team, Belen Alastruey, Niyati Bafna, Andrea Caciolai, Kevin Heffernan, Artyom Kozhevnikov, Christophe Ropers, Eduardo Sánchez, Charles-Eric Saint-James, Ioannis Tsiamas, Chierh CHENG, Joe Chuang, Paul-Ambroise Duquenne, Mark Duppenthaler, Nate Ekberg, Cynthia Gao, Pere Lluís Huguet Cabot, João Maria Janeiro, Jean Maillard, Gabriel Mejia Gonzalez, Holger Schwenk, Edan Toledo, Arina Turkatenko, Albert Ventayol-Boada, Rashel Moritz, Alexandre Mourachko, Surya Parimi, Mary Williamson, Shireen Yates, David Dale, Marta R. Costa-jussa</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Omnilingual SONAR: Cross-Lingual and Cross-Modal Sentence Embeddings Bridging Massively Multilingual Text and Speech</title>
<description>Cross-lingual sentence encoders have traditionally been limited to a few hundred languages, and have sacrificed downstream performance to achieve better alignment across languages, limiting their adoption. In this work, we introduce OmniSONAR, a novel family of omnilingual, cross-lingual and cross-modal sentence embedding models that breaks this barrier. We establish a unified semantic space, natively encompassing text, speech, code and mathematical expressions, while achieving state-of-the-art downstream performance for an unprecedented scale of thousands of languages, from high-resource languages to extremely low-resource varieties.
To achieve this scale without representation collapse and while maintaining top-tier performance in the high-resource languages, we employ a progressive training strategy. We first build a state-of-the-art foundational embedding space for 200 languages using an LLM-initialized Encoder-Decoder, combining token-level decoding with a novel split-softmax contrastive loss and synthetic hard negatives. Leveraging this strong foundational space, we expand to several thousands of language varieties via a specialized two-stage teacher-student encoder distillation framework. Further modeling extensions derived from OmniSONAR address long context inputs and token-centric representations. Finally, we demonstrate the cross-modal extensibility of this space by seamlessly mapping 177 spoken languages into it.
OmniSONAR redefines the state of the art for multilingual representation learning. It halves the cross-lingual similarity search error rate of the previous best models on the 200 languages of FLORES, while also achieving a staggering 15-fold error rate reduction across 1,560 languages in the BIBLE benchmark. Furthermore, our embedding model enables unprecedented translation capabilities, outperforming NLLB-3B on several multilingual benchmarks, and surpassing all previous models, including multi-billion-parameter LLMs, by 15 chrF++ points in 1,560→English translation in the BIBLE benchmark. Beyond alignment and translation, OmniSONAR demonstrates strong general-purpose capabilities across downstream embedding tasks on MTEB and programming languages on XLCoST. For the speech modality, our massively multilingual extension exhibits a 43% lower error rate in cross-lingual and cross-modal similarity search, while achieving 97% of SeamlessM4T performance in speech-to-text translation, despite being a zero-shot translation model trained only with ASR data. Finally, by training an encoder-decoder language model, Spectrum, exclusively on English text that processes OmniSONAR sequences, we unlock immediate high-performance transfer to thousands of languages and the speech modality for complex downstream tasks. These outstanding results position OmniSONAR as a robust, language- and modality-agnostic foundation for any downstream usage.</description>
<link>https://ai.meta.com/research/publications/omnilingual-sonar-cross-lingual-and-cross-modal-sentence-embeddings-bridging-massively-multilingual-text-and-speech/</link>
<guid isPermaLink="false">1231952259116893</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual SONAR Team, João Maria Janeiro, Pere Lluís Huguet Cabot, Ioannis Tsiamas, Yen Meng, Vivek Iyer, Guillem Ramirez, Loic Barrault, Belen Alastruey, Yu-An Chung, Marta R. Costa-jussa, David Dale, Kevin Heffernan, Jaehyeong Jo, Artyom Kozhevnikov, Alexandre Mourachko, Christophe Ropers, Holger Schwenk, Paul-Ambroise Duquenne</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Speech & Audio</category>
</item>
<item>
<title>Four MTIA Chips in Two Years: Scaling AI Experiences for Billions</title>
<description>Serving a wide range of AI models on a global scale, while maintaining the lowest possible costs, is one of the most demanding infrastructure challenges in the industry. </description>
<link>https://ai.meta.com/blog/meta-mtia-scale-ai-chips-for-billions/</link>
<guid isPermaLink="false">814227801000565</guid>
<pubDate>Wed, 11 Mar 2026 14:00:55 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/649719916_2147323009439466_6866394022243367245_n.png?_nc_cat=103&ccb=1-7&_nc_sid=e280be&_nc_ohc=HQUtIT-12s8Q7kNvwHypiBd&_nc_oc=AdqhDC5yIofMP9r2YvlWFr93_aG4GuJ3-iTW5khRJZ89bZxZNaauqtdJK00B6-15sWs&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af3rww88qVhnDDRUH8TNmMCPOAbmOj46CJxWcclu6JbCgw&oe=6A08CB6C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Mapping the World's Forests with Greater Precision: Introducing Canopy Height Maps v2</title>
<description>In partnership with the World Resources Institute, today we’re announcing Canopy Height Maps v2 (CHMv2), an open source model, along with world-scale maps generated with the model.</description>
<link>https://ai.meta.com/blog/world-resources-institute-dino-canopy-height-maps-v2/</link>
<guid isPermaLink="false">1675459620286410</guid>
<pubDate>Tue, 10 Mar 2026 16:00:06 GMT</pubDate>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.2365-6/644822799_1500169394861918_3213642176014981812_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=5tvZOejkl8MQ7kNvwGvu4MO&_nc_oc=AdrVs4z8wgcKxzuHwX-nVRZ2jwpvpcXnCoqTV1dyBz47sxm5tkNoYpwnT1Jj3LcN8fY&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af06GeDgtcDeenbHjQihlQDmqGWyV0jynSOZy2vqB6xJuw&oe=6A08D606" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
<item>
<title>Unified Vision–Language Modeling via Concept Space Alignment</title>
<description>We introduce v-Sonar, a vision–language embedding space extended from the text-only embedding space Sonar (Omnilingual Embeddings Team et al., 2026), which supports 1500 text languages and 177 speech languages. To construct v-Sonar, we propose a post-hoc alignment pipeline that maps the representations of an existing vision encoder into the Sonar space. We thoroughly evaluate v-Sonar and show that its embeddings achieve competitive performance on text-to-video retrieval. Equipped with the Sonar text decoder, v-Sonar further surpasses state-of-the-art vision–language models on video captioning tasks, including Dream-1k (Bleu 24.3 vs. 19.6) and Vatex (Bleu 45.0 vs. 41.5).
Leveraging v-Sonar, we first demonstrate that the Large Concept Model (LCM; LCM team et al. 2024) operating in Sonar and trained with English text only, can perform both single- and multi-visual concept understanding in a zero-shot manner. Finally, we introduce v-LCM, which extends the LCM with vision–language instruction tuning. v-LCM encodes vision and language inputs into an unified sequence of latent embeddings via v-Sonar and Sonar, and it is trained with the same latent diffusion objective for next-embedding prediction as in LCM’s text-only pre-training. Experiments on a large-scale multilingual and -modal instruction–tuning data mixture highlight the potential of v-LCM: v-LCM matches state-of-the-art vision-language models on tasks covering image/video captioning and question answering, while significantly outperforming them across 61 rich- to low-resource languages out of all 62 tested languages.</description>
<link>https://ai.meta.com/research/publications/unified-vision-language-modeling-via-concept-space-alignment/</link>
<guid isPermaLink="false">1257213679668978</guid>
<pubDate>Fri, 27 Feb 2026 08:00:00 GMT</pubDate>
<author>Yifu Qiu, Paul-Ambroise Duquenne, Holger Schwenk</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Research</category>
</item>
<item>
<title>Learning Personalized Agents from Human Feedback</title>
<description>Modern AI agents are powerful but often fail to align with the idiosyncratic, evolving preferences of individual users. Prior approaches typically rely on static datasets, either training implicit preference models on interaction history or encoding user profiles in external memory. However, these approaches struggle with new users and with preferences that change over time. We introduce Personalized Agents from Human Feedback (PAHF), a framework for continual personalization in which agents learn online from live interaction using explicit per-user memory. PAHF operationalizes a three-step loop: (1) seeking pre-action clarification to resolve ambiguity, (2) grounding actions in preferences retrieved from memory, and (3) integrating post-action feedback to update memory when preferences drift. To evaluate this capability, we develop a four-phase protocol and two benchmarks in embodied manipulation and online shopping. These benchmarks quantify an agent's ability to learn initial preferences from scratch and subsequently adapt to persona shifts. Our theoretical analysis and empirical results show that integrating explicit memory with dual feedback channels is critical: PAHF learns substantially faster and consistently outperforms both no-memory and single-channel baselines, reducing initial personalization error and enabling rapid adaptation to preference shifts.</description>
<link>https://ai.meta.com/research/publications/learning-personalized-agents-from-human-feedback/</link>
<guid isPermaLink="false">910245944695813</guid>
<pubDate>Thu, 26 Feb 2026 08:00:00 GMT</pubDate>
<author>Kaiqu Liang, Julia Kruk, Shengyi Qian, Xianjun Yang, Shengjie Bi, Shaoliang Nie, Michael Zhang, Lijuan Liu, Jaime Fernández Fisac, Shuyan Zhou, Saghar Hosseini</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Conversational AI</category>
<category>Research</category>
</item>
<item>
<title>FERRET: Framework for Expansion Reliant Red Teaming</title>
<description>We introduce a multi-faceted automated red teaming framework in which the goal is to generate multi-modal adversarial conversations that would break a target model and introduce various expansions that would result in more effective and efficient adversarial conversations. The introduced expansions include: 1. Horizontal expansion in which the goal is for the red team model to self-improve and generate more effective conversation starters that would shape a conversation. 2. Vertical expansion in which the goal is to take these conversation starters that are discovered in the horizontal expansion phase and expand them into effective multi-modal conversations and 3. Meta expansion in which the goal is for the red team model to discover more effective multi-modal attack strategies during the course of a conversation. We call our framework FERRET (Framework for Expansion Reliant Red Teaming) and compare it with various existing automated red teaming approaches. In our experiments, we demonstrate the effectiveness of FERRET in generating effective multi-modal adversarial conversations and its superior performance against existing state of the art approaches.</description>
<link>https://ai.meta.com/research/publications/ferret-framework-for-expansion-reliant-red-teaming/</link>
<guid isPermaLink="false">2396003004185464</guid>
<pubDate>Fri, 13 Feb 2026 08:00:00 GMT</pubDate>
<author>Ninareh Mehrabi, Vítor Albiero, Maya Pavlova, Joanna Bitton</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Responsible AI</category>
</item>
<item>
<title>UniT: Unified Multimodal Chain-of-Thought Test-time Scaling</title>
<description>Unified models can handle both multimodal understanding and generation within a single architecture, yet they typically operate in a single pass without iteratively refining their outputs. Many multimodal tasks, especially those involving complex spatial compositions, multiple interacting objects, or evolving instructions, require decomposing instructions, verifying intermediate results, and making iterative corrections. While test-time scaling (TTS) has demonstrated that allocating additional inference compute for iterative reasoning substantially improves language model performance, extending this paradigm to unified multimodal models remains an open challenge.
We introduce UniT, a framework for multimodal chain-of-thought test-time scaling that enables a single unified model to reason, verify, and refine across multiple rounds. UniT combines agentic data synthesis, unified model training, and flexible test-time inference to elicit cognitive behaviors including verification, subgoal decomposition, and content memory. Our key findings are: (1) unified models trained on short reasoning trajectories generalize to longer inference chains at test time; (2) sequential chain-of-thought reasoning provides a more scalable and compute-efficient TTS strategy than parallel sampling; (3) training on generation and editing trajectories improves out-of-distribution visual reasoning. These results establish multimodal test-time scaling as an effective paradigm for advancing both generation and understanding in unified models.</description>
<link>https://ai.meta.com/research/publications/unit-unified-multimodal-chain-of-thought-test-time-scaling/</link>
<guid isPermaLink="false">1477019817147720</guid>
<pubDate>Wed, 11 Feb 2026 08:00:00 GMT</pubDate>
<author>Leon Liangyu Chen, Haoyu Ma, Zhipeng Fan, Ziqi Huang, Animesh Sinha, Xiaoliang Dai, Jialiang Wang, Zecheng He, Jianwei Yang, Chunyuan Li, Junzhe Sun, Chu Wang, Serena Yeung-Levy, Felix Juefei-Xu</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Computer Vision</category>
</item>
<item>
<title>AIRS-Bench: a Suite of Tasks for Frontier AI Research Science Agents</title>
<description>LLM agents hold significant promise for advancing scientific research. To accelerate this progress, we introduce AIRS-Bench (the AI Research Science Benchmark), a suite of 20 tasks sourced from state-of-the-art machine learning papers. These tasks span diverse domains, including language modeling, mathematics, bioinformatics, and time series forecasting. AIRS-Bench tasks assess agentic capabilities over the full research lifecycle---including idea generation, experiment analysis and iterative refinement---without providing baseline code. The AIRS-Bench task format is versatile, enabling easy integration of new tasks and rigorous comparison across different agentic frameworks. We establish baselines using frontier models paired with both sequential and parallel scaffolds. Our results show that agents exceed human SOTA in four tasks but fail to match it in sixteen others. Even when agents surpass human benchmarks, they do not reach the theoretical performance ceiling for the underlying tasks. These findings indicate that AIRS-Bench is far from saturated and offers substantial room for improvement.
We open-source the AIRS-Bench task definitions and evaluation code to catalyze further development in autonomous scientific research.</description>
<link>https://ai.meta.com/research/publications/airs-bench-a-suite-of-tasks-for-frontier-ai-research-science-agents/</link>
<guid isPermaLink="false">2685677328474642</guid>
<pubDate>Tue, 10 Feb 2026 08:00:00 GMT</pubDate>
<author>Alisia Lupidi, Bhavul Gauri, Thomas Simon Foster, Bassel Al Omari, Despoina Magka, Alberto Pepe, Alexis Audran-Reiss, Muna Aghamelu, Nicolas Baldwin, Lucia Cipolina-Kun, Jean-Christophe Gagnon-Audet, Chee Hau Leow, Sandra Lefdal, Hossam Mossalam, Abhinav Moudgil, Saba Nazir, Emanuel Tewolde, Isabel Urrego, Jordi Armengol-Estape, Amar Budhiraja, Gaurav Chaurasia, Abhishek Charnalia, Derek Dunfield, Karen Hambardzumyan, Daniel Izcovich, Martin Josifoski, Ishita Mediratta, Kelvin Niu, Parth Pathak, Michael Shvartsman, Edan Toledo, Anton Protopopov, Roberta Raileanu, Alexander Miller, Tatiana Shavrina, Jakob Foerster, Yoram Bachrach</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0QjqpFsFaKTQsPJaXp3nUDapnuR66fyethTLb73gqNCA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
</item>
<item>
<title>Reducing Government Costs and Increasing Access to Greenspaces in the United Kingdom with DINO</title>
<description>Meta's DINOv2 model is enhancing reforestation efforts around the world. Learn how the UK government is using DINO to help reduce costs and increase access to greenspaces.</description>
<link>https://ai.meta.com/blog/forest-research-dino/</link>
<guid isPermaLink="false">1369350751606207</guid>
<pubDate>Mon, 09 Feb 2026 17:25:45 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/631673937_4911145895833160_8225979506389394021_n.gif?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=N94hGWEZkvUQ7kNvwGqt8rU&_nc_oc=AdokV6EWG3vMd59CgFsHxXNMWZyGwZg3dVfws-Jhaxx2eEfaisVJ7EZypbfXY18vfqk&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0hIpDvBB4Bnp6Cc_yqaNnc2eUmoWEiCH-bpj4NLgTvxA&oe=6A08CAFF" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
<category>Open Source</category>
</item>
<item>
<title>Multi-room Apartments Simulation (MRAS) Dataset | Meta AI Research</title>
<description>The Multi-Room Apartments Simulation (MRAS) dataset is a multi-modal dataset created for the task of estimating spatially-distributed acoustic parameters in complex scenes.</description>
<link>https://ai.meta.com/datasets/mras-simulated-rooms-dataset/</link>
<guid isPermaLink="false">2247220155683256</guid>
<pubDate>Wed, 28 Jan 2026 18:49:18 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/612410014_2220471548441535_606775622330495537_n.jpg?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=1bE1l45d6ZQQ7kNvwGMtrFY&_nc_oc=Adqxxb2bGLw1oq3NG8aZYuMODgbpt__dmsQw5pkrsOQbRgeHEIStu0h09rsY2lXjCgg&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af2zeJMEn_1sA21ibQRd4o7aQTDv_8ZxVJtR6xarCJs81A&oe=6A08CB19" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
<item>
<title>How DINO and SAM are Helping Modernize Essential Medical Triage Practices</title>
<description>By leveraging advanced AI models, teams at the University of Pennsylvania are aiming to bring cutting-edge automation to emergency response.</description>
<link>https://ai.meta.com/blog/upenn-dino-sam-helping-medical-triage/</link>
<guid isPermaLink="false">3255814814579669</guid>
<pubDate>Thu, 18 Dec 2025 18:00:20 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/597680380_857278207264023_6561532616114241333_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=2MKpeQyaIXMQ7kNvwEW_JOV&_nc_oc=Adoz5nmhj_wrxp1GCBlKBrGXATQAKT54gCSnbUJwgRtuQvy7TBeFM8JWoOf4NqzAroI&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af2o_TRAw1BFszWN99x133m-mBr4WFubFhdt55DnQBmmKw&oe=6A08D437" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>The Universities Space Research Association Applies Segment Anything Model for Responding to Flood Emergencies</title>
<description>The Universities Space Research Association and Meta are collaborating to help support water observing systems set up by the U.S. Geological Survey. </description>
<link>https://ai.meta.com/blog/usra-sam-flood-emergencies/</link>
<guid isPermaLink="false">1381613153671193</guid>
<pubDate>Thu, 18 Dec 2025 18:00:12 GMT</pubDate>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.2365-6/600304853_1519099669397813_6606203512792368025_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=pBY1dQ-wuD4Q7kNvwEtk3Er&_nc_oc=AdqCTI1KUCH0R0iK0pnBK-3F96nsHIHi7HyQscda0R_SXolcNLBmVeOzZ5XnS2bqPMw&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af1wBHNFgQlmiXdmBPOYQkcZ462G1bhx4kreni9KeHzLXQ&oe=6A08B857" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>How Orakl Oncology is using DINOv2 to accelerate cancer treatment discovery</title>
<description>Orakl Oncology aims to accelerate cancer research and drug development by combining experimental, lab-based insights and machine learning.</description>
<link>https://ai.meta.com/blog/orakl-oncology-dinov2-accelerating-cancer-treatment/</link>
<guid isPermaLink="false">1035491295071881</guid>
<pubDate>Thu, 20 Feb 2025 17:00:31 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/480460718_649637170770478_3994719197111548806_n.jpg?_nc_cat=106&ccb=1-7&_nc_sid=e280be&_nc_ohc=FE_ybvE6IoMQ7kNvwHxpHqx&_nc_oc=AdqKWcB-oYsR3a2eNDId3y_5qtXGIqXAUknFknymdM_3m5mA8aYdrL9y6E2-1jNZVyw&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af1GIPCQrtDHPZV0nuzVNIUW2uL4KULa935Rjvn4V_XdhQ&oe=6A08C327" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
<item>
<title>Aaron Defazio</title>
<description>Aaron's research focuses on improving the practice of machine learning through the development of more reliable and theoretically sound methods such as ...</description>
<link>https://ai.meta.com/people/1115638629589333/aaron-defazio/</link>
<guid isPermaLink="false">1115638629589333</guid>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.35477-6/431543048_2088159024878120_1553708304445833368_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=109&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=vd7kajaxkn4Q7kNvwGQ7l84&_nc_oc=AdrVzy9eg4TqCHzxcCZ1L4XI2I740RBa3s9PgGhl0KVDLXIe-YZTwbQUpKKl868Ecec&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0lotCliC_qi23JFbDSoU_I5pLpD9LNg3C4tpNHGJC2tA&oe=69F449C8" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Theory</category>
<category>Core Machine Learning</category>
<category>Research</category>
<category>Speech & Audio</category>
<category>NLP</category>
</item>
<item>
<title>Abhimanyu Dubey</title>
<description>Abhimanyu is a Research Scientist at Facebook AI Research working on problems in machine learning. He received his Ph.D. in 2021 from the Massachusetts ...</description>
<link>https://ai.meta.com/people/1401136157273652/abhimanyu-dubey/</link>
<guid isPermaLink="false">1401136157273652</guid>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.35477-6/433945609_1344738226210370_2541979406521528116_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=54biatSobnEQ7kNvwHm-noQ&_nc_oc=AdqgN3XWPgzN9zeEi07MRfAbFtwvMz1HwIPWMkl_i0ZncDiOKPKCljvu2nsTH9e6brE&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af22EpGZawm0UsFBEgVqtyM5pTOK2IWYeLhaytkZSmaSKA&oe=69F4610D" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Computer Vision</category>
<category>Research</category>
<category>ML Applications</category>
<category>Human & Machine Intelligence</category>
<category>Core Machine Learning</category>
</item>
<item>
<title>Abhishek Das</title>
<description>Abhishek is a Research Scientist at Facebook AI Research (FAIR). His research focuses on deep learning and its applications in climate change, and in bu...</description>
<link>https://ai.meta.com/people/401335699167238/abhishek-das/</link>
<guid isPermaLink="false">401335699167238</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431537252_1543342826243118_6010155505132587328_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=102&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=7bKy_GgNvJUQ7kNvwHnc-io&_nc_oc=AdqcGdwhDx2dXwHqsEc4_GeDLVyC-LnWdC1bV1-bWeL-N8jcQe-nKtJ30FPvIdT16SM&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af2x6fwzH_pvuxnwQ20rUPtee0Fw3HMm8qJpOTjEW6Wrvg&oe=69F44FDB" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
<category>Computer Vision</category>
<category>NLP</category>
</item>
<item>
<title>Adam Polyak</title>
<description>Adam is a research engineer at Facebook AI Research (FAIR), focusing on deep learning, cross domain image generation, audio generation, and voice conver...</description>
<link>https://ai.meta.com/people/1434021433901222/adam-polyak/</link>
<guid isPermaLink="false">1434021433901222</guid>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.35477-6/431518452_1124340545357794_1947310908456905622_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=111&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=qnsH_8koyHsQ7kNvwFb8I9k&_nc_oc=AdqF4anURrXM4aReFHnW4Car9JsZOksY4Hr5eqp9BLYAY_vkyIocVQKUTHBgQv8u84c&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af3rwXnrgI2OK5qHEom1Je1cPLcsT5NhIdACX5COiUW1Dg&oe=69F4377A" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Research</category>
<category>ML Applications</category>
<category>Speech & Audio</category>
</item>
<item>
<title>Adeel Cheema</title>
<description>Adeel is a software engineer working on transparency &amp; control within Meta’s Responsible AI pillar. He focuses on creating product experiences to expl...</description>
<link>https://ai.meta.com/people/1808244339639715/adeel-cheema/</link>
<guid isPermaLink="false">1808244339639715</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431551867_767017845524520_8549663744281985032_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=3dHbSn4toEQQ7kNvwG-hxyL&_nc_oc=AdrneKtcRtDVD2x9eU9c0e9lX-OK623WzkPRI08JRJ3HANhfIiPoSRx5_FjSLcFEWTg&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af1uwPhfwYIXUK7E8qNdWDO_0w35oIC9m2Lpepe4LaNoLA&oe=69F44D79" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Adina Williams</title>
<description>Adina is a Research Scientist at Facebook AI Research in NYC (started October 2018). Previously, she earned her PhD at New York University in the Depart...</description>
<link>https://ai.meta.com/people/1396973444287406/adina-williams/</link>
<guid isPermaLink="false">1396973444287406</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431521840_2635057386653799_4484141047373535802_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=104&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=D9tP58vAF5kQ7kNvwHvASYA&_nc_oc=AdrKJZSIAfBDaQnc-wW1ToZsOSEhxJpnizTtfwAZJHHOlkTEt1y7LNyQXm2jzG7quUg&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af0x9ldbeM0upEUiAofAAtTc6p0WCZnp9iRLG1D-YF_qfQ&oe=69F467EC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Computer Vision</category>
<category>Core Machine Learning</category>
<category>Open Source</category>
<category>NLP</category>
<category>Conversational AI</category>
<category>Human & Machine Intelligence</category>
<category>ML Applications</category>
</item>
<item>
<title>Adriana Romero Soriano</title>
<description>Adriana is currently a research scientist at Facebook AI Research and an adjunct professor at McGill University. The goal of her research is to develop ...</description>
<link>https://ai.meta.com/people/425710613281765/adriana-romero-soriano/</link>
<guid isPermaLink="false">425710613281765</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431577893_1361257101250742_6288541672250118710_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=nsGSoSWY8rkQ7kNvwFMCuJe&_nc_oc=AdpfuVWkMsXwut0OSpmRwcFwPnozYpICadHJKwLA54PhHlVuRaKtTNxBtL3TcbLP6hk&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af3lo96TxPCS1iEIBmDuCLBlMDU5FVjbLkdT0RDjqEuHtQ&oe=69F44DCC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Core Machine Learning</category>
<category>Conversational AI</category>
<category>Reinforcement Learni9ng</category>
</item>
<item>
<title>Akhil Mathur</title>
<description>Akhil Mathur is a Research Scientist in the Llama Research team at Meta, working on enhancing the reasoning abilities of Llama models. Previously, as a ...</description>
<link>https://ai.meta.com/people/1251205439201018/akhil-mathur/</link>
<guid isPermaLink="false">1251205439201018</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/457008091_844600247808930_1170475198447093629_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=A2xSui7yUHAQ7kNvwHWOVVV&_nc_oc=Adp4eRq3NJ0287FNM_lETOYqZPQRg2pdbU9cuGT_PdgE67VDPw_dwbYD26U2CO3__R8&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af2hqqM4Gpjxcuda6JgfAcVcdZ4C7PznBqjPC2Ozql1p-A&oe=69F4499C" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Akshara Rai</title>
<description>Akshara is a Research Scientist at Facebook AI Research, working at the intersection of machine learning and control. Her research aims at teaching robo...</description>
<link>https://ai.meta.com/people/423502423531745/akshara-rai/</link>
<guid isPermaLink="false">423502423531745</guid>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.35477-6/433923962_1381208202581063_6502596142308749692_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=0Wm1dG_MC68Q7kNvwGTCKdY&_nc_oc=AdoD6CrJ1m76HwlDd53FFb_r2h87MuGA8lBXoe2rT1bAb0Kfh-5LqBXoWovSqruL6lU&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af2k5nxfcwtZ3RYYOJIZc_NFFVmkLdi0M2GrrlN9PPyi5Q&oe=69F43F4F" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Robotics</category>
<category>Computer Vision</category>
<category>Research</category>
<category>Reinforcement Learni9ng</category>
</item>
<item>
<title>Alborz Geramifard</title>
<description>Alborz Geramifard is a senior research manager at Meta AI supporting the Conversational AI. Prior to joining Meta, he led the conversational AI team at ...</description>
<link>https://ai.meta.com/people/363544420007153/alborz-geramifard/</link>
<guid isPermaLink="false">363544420007153</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431495520_3457483597876936_681607123433531609_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=QzBw61zVxHEQ7kNvwHFGECJ&_nc_oc=Adp_dD05XDi2euaYM6c0WcWx7YkqPn1LQ4QiGQnKLMbD88ygYqi7HyV0szo-h1D2zqM&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af19ZQy5OUEeoNdYhiV4auQwhX7LA-Rz7UWd0YXB0Sdgpw&oe=69F45A51" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
<category>Conversational AI</category>
<category>NLP</category>
<category>Computer Vision</category>
</item>
<item>
<title>Alexander H. Miller</title>
<description>Alexander has worked in multi-domain research engineering management, supporting research engineers in London, Tel Aviv, and New York. He is advanced in...</description>
<link>https://ai.meta.com/people/264624513377272/alexander-h-miller/</link>
<guid isPermaLink="false">264624513377272</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/433950826_944833516851120_841662693953017766_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=6ftFVCshb-8Q7kNvwFmW3Vx&_nc_oc=AdpYZTkVT_RFDrpt-_sscuIcqOYLysabf-d4BJfV3EnXsHNOYkP_SRmSHFKxebGWjaU&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af3k9H3xSidV17ChG_HQ25x_UTsic22d0dR3Y4_MXOqutQ&oe=69F45C6C" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Reinforcement Learni9ng</category>
<category>Graphics</category>
<category>NLP</category>
</item>
<item>
<title>Amey Porobo Dharwadker</title>
<description>Amey Porobo Dharwadker works as a Machine Learning Engineering Manager, leading Facebook's Video Recommendations Ranking team. Renowned for his expertis...</description>
<link>https://ai.meta.com/people/393924029921120/amey-porobo-dharwadker/</link>
<guid isPermaLink="false">393924029921120</guid>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.35477-6/431510720_7490594577724788_4289788014840147933_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=106&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=CDFvi0iDZRcQ7kNvwFhizrj&_nc_oc=Adr9sJsLUd39ruFsTcvrZXX1i_tYXLf-EyYjs5A_aQ4ACUVT9oNSSPcZpMVShrui3pk&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=ZsOxLez5mt9lNLZN1lyOGA&_nc_ss=73289&oh=00_Af1HhDHjFJzIEU7lQZ6kLL4col4ANeeMcxQUU6E7f250eg&oe=69F440F5" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
</channel>
</rss>... |
github-actions
Bot
added
the
auto: ready to review
Contributor
http://localhost:1200/meta/ai/global-search/content_types=blog - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — content_types=blog</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/content_types=blog" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/ (sort: RELEVANCE, total hits: 589). - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:09:47 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>Scaling How We Build and Test Our Most Advanced AI</title>
<description>As we build more capable, personalized AI, reliability, security, and user protections are more important than ever. </description>
<link>https://ai.meta.com/blog/scaling-how-we-build-test-advanced-ai/</link>
<guid isPermaLink="false">4369469056623958</guid>
<pubDate>Wed, 08 Apr 2026 17:00:17 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/668130848_1453339552354697_3291493920555502062_n.png?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=yvFttOBoRYkQ7kNvwE2zAxR&_nc_oc=Adr2UifKLIVIje1oDq_d_Zzzd2xhHfN26iDMLvatYMqMAKMUrJtd8QiHDdByuQsgxo8&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af1eIy9_tG4me9_FWK0M-JHgfqG16AGKafOxuYaVPEjKdQ&oe=6A08CDEE" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Introducing Muse Spark: Scaling Towards Personal Superintelligence</title>
<description></description>
<link>https://ai.meta.com/blog/introducing-muse-spark-msl/</link>
<guid isPermaLink="false">963191949487556</guid>
<pubDate>Wed, 08 Apr 2026 16:00:59 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/667507363_1871749950191919_4996140970371485757_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=IbntS1cBAcsQ7kNvwFGpZpf&_nc_oc=AdoWRIZVlnRJCHKphpg-44LJdWpIcGyseuyTtGVMwVPvYn8IC_2M7OqrJOG1JaQJSk0&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af2wTnmlBOrpykYuhRvEoQipmyYyw-X51kfEm4QvSvraVw&oe=6A08D22C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>How Alta Daily Uses Meta’s Segment Anything to Reimagine the Digital Closet</title>
<description></description>
<link>https://ai.meta.com/blog/alta-daily-fashion-app-segment-anything/</link>
<guid isPermaLink="false">2194788291268065</guid>
<pubDate>Mon, 06 Apr 2026 18:53:27 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/658967078_993415586351079_325919381939729048_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=cBKF2Wk0XYoQ7kNvwHg8n78&_nc_oc=Adpp8hWFe9GVSkrNsSGgcTK4r7FP6jRCnFsR4SVl4s8_NG3Td8pcFrQWplQHD9AJ0_Y&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af1AIT-oJ9kafNaaoqRM-XmzZaLXRW3qxDHTTOpmNLcUGw&oe=6A08B252" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning</title>
<description></description>
<link>https://ai.meta.com/blog/segment-anything-model-3/</link>
<guid isPermaLink="false">715047027731045</guid>
<pubDate>Fri, 27 Mar 2026 16:00:37 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/583339165_852765810541557_5592130597432767690_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=WvtkkpGzh44Q7kNvwEubR7-&_nc_oc=AdosYwdove82TFwfnCLk3q5bXi6OYKYKFlbayU_okqqiX8vtIP70WBbgYw_OqvZ58J8&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af1KQgIkEGastrUp1amzBVnUAkgHz_E7trgGfjccFIKehw&oe=6A08D061" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>Introducing TRIBE v2: A Predictive Foundation Model Trained to Understand How the Human Brain Processes Complex Stimuli</title>
<description></description>
<link>https://ai.meta.com/blog/tribe-v2-brain-predictive-foundation-model/</link>
<guid isPermaLink="false">25512119895129580</guid>
<pubDate>Thu, 26 Mar 2026 16:00:16 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/659112495_952140057346056_1728296311373871062_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=rBe_VZxcFD4Q7kNvwGOhdP7&_nc_oc=AdoaYP1BwQzfnCGpSQaVG8V8CaMbbznHgw9z7Mt3CIaeiUr0iUJpZ_p5I6em3s7sWpk&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af33glGLnMAPNr6DNjU5i_O5GfZryVsdt6TyipyurRsFIg&oe=6A08AF6B" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Research</category>
</item>
<item>
<title>Four MTIA Chips in Two Years: Scaling AI Experiences for Billions</title>
<description>Serving a wide range of AI models on a global scale, while maintaining the lowest possible costs, is one of the most demanding infrastructure challenges in the industry. </description>
<link>https://ai.meta.com/blog/meta-mtia-scale-ai-chips-for-billions/</link>
<guid isPermaLink="false">814227801000565</guid>
<pubDate>Wed, 11 Mar 2026 14:00:55 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/649719916_2147323009439466_6866394022243367245_n.png?_nc_cat=103&ccb=1-7&_nc_sid=e280be&_nc_ohc=HQUtIT-12s8Q7kNvwHypiBd&_nc_oc=AdqhDC5yIofMP9r2YvlWFr93_aG4GuJ3-iTW5khRJZ89bZxZNaauqtdJK00B6-15sWs&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af3FCW3N1Ly8sLHB9Kmh_hPhSbvXBpjz8IjVo7IYiWNXcw&oe=6A08CB6C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Mapping the World's Forests with Greater Precision: Introducing Canopy Height Maps v2</title>
<description>In partnership with the World Resources Institute, today we’re announcing Canopy Height Maps v2 (CHMv2), an open source model, along with world-scale maps generated with the model.</description>
<link>https://ai.meta.com/blog/world-resources-institute-dino-canopy-height-maps-v2/</link>
<guid isPermaLink="false">1675459620286410</guid>
<pubDate>Tue, 10 Mar 2026 16:00:06 GMT</pubDate>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.2365-6/644822799_1500169394861918_3213642176014981812_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=5tvZOejkl8MQ7kNvwGvu4MO&_nc_oc=AdrVs4z8wgcKxzuHwX-nVRZ2jwpvpcXnCoqTV1dyBz47sxm5tkNoYpwnT1Jj3LcN8fY&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af18hzBaDnN9W2nGHlVlhTZGlsqsepVuMdSUIGe1_z7noA&oe=6A08D606" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
<item>
<title>Reducing Government Costs and Increasing Access to Greenspaces in the United Kingdom with DINO</title>
<description>Meta's DINOv2 model is enhancing reforestation efforts around the world. Learn how the UK government is using DINO to help reduce costs and increase access to greenspaces.</description>
<link>https://ai.meta.com/blog/forest-research-dino/</link>
<guid isPermaLink="false">1369350751606207</guid>
<pubDate>Mon, 09 Feb 2026 17:25:45 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/631673937_4911145895833160_8225979506389394021_n.gif?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=N94hGWEZkvUQ7kNvwGqt8rU&_nc_oc=AdokV6EWG3vMd59CgFsHxXNMWZyGwZg3dVfws-Jhaxx2eEfaisVJ7EZypbfXY18vfqk&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af1uhS_3iTRditSn2lN89MQkPMFun4LXLS2Gzb1lnasLSA&oe=6A08CAFF" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
<category>Open Source</category>
</item>
<item>
<title>How DINO and SAM are Helping Modernize Essential Medical Triage Practices</title>
<description>By leveraging advanced AI models, teams at the University of Pennsylvania are aiming to bring cutting-edge automation to emergency response.</description>
<link>https://ai.meta.com/blog/upenn-dino-sam-helping-medical-triage/</link>
<guid isPermaLink="false">3255814814579669</guid>
<pubDate>Thu, 18 Dec 2025 18:00:20 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/597680380_857278207264023_6561532616114241333_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=2MKpeQyaIXMQ7kNvwEW_JOV&_nc_oc=Adoz5nmhj_wrxp1GCBlKBrGXATQAKT54gCSnbUJwgRtuQvy7TBeFM8JWoOf4NqzAroI&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af3Tjm7aqm7nNnckgW0fTL3BtEmNP09eVOkcw70MzyrNFQ&oe=6A08D437" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>The Universities Space Research Association Applies Segment Anything Model for Responding to Flood Emergencies</title>
<description>The Universities Space Research Association and Meta are collaborating to help support water observing systems set up by the U.S. Geological Survey. </description>
<link>https://ai.meta.com/blog/usra-sam-flood-emergencies/</link>
<guid isPermaLink="false">1381613153671193</guid>
<pubDate>Thu, 18 Dec 2025 18:00:12 GMT</pubDate>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.2365-6/600304853_1519099669397813_6606203512792368025_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=pBY1dQ-wuD4Q7kNvwEtk3Er&_nc_oc=AdqCTI1KUCH0R0iK0pnBK-3F96nsHIHi7HyQscda0R_SXolcNLBmVeOzZ5XnS2bqPMw&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af24LDwNJffGMD-gBfveOAvcJepJ_gZfH9cyjlJsGTY2bQ&oe=6A08B857" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>Introducing SAM Audio: The First Unified Multimodal Model for Audio Separation</title>
<description>SAM Audio transforms audio processing by making it easy to isolate any sound from complex audio mixtures using natural, multimodal prompts — whether through text, visual cues, or marking time segments. </description>
<link>https://ai.meta.com/blog/sam-audio/</link>
<guid isPermaLink="false">1658975421738456</guid>
<pubDate>Tue, 16 Dec 2025 18:00:17 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/600517835_858099609958067_2071268735982577268_n.png?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=jNAx0Vnb61IQ7kNvwGRiU6v&_nc_oc=Ado1fsJq-6PbHjCC_C4kXWI17LkRHZuDdofXBRv5tFmz7Ws1r29R-G_r4qipxlZd7sE&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af2MrNw7mScGE3MH9Xbni_h5Uwu5GKUAxb7jd_ZFJnHXkw&oe=6A08B61B" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>How Orakl Oncology is using DINOv2 to accelerate cancer treatment discovery</title>
<description>Orakl Oncology aims to accelerate cancer research and drug development by combining experimental, lab-based insights and machine learning.</description>
<link>https://ai.meta.com/blog/orakl-oncology-dinov2-accelerating-cancer-treatment/</link>
<guid isPermaLink="false">1035491295071881</guid>
<pubDate>Thu, 20 Feb 2025 17:00:31 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/480460718_649637170770478_3994719197111548806_n.jpg?_nc_cat=106&ccb=1-7&_nc_sid=e280be&_nc_ohc=FE_ybvE6IoMQ7kNvwHxpHqx&_nc_oc=AdqKWcB-oYsR3a2eNDId3y_5qtXGIqXAUknFknymdM_3m5mA8aYdrL9y6E2-1jNZVyw&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=boSyHDhF5v2sPu82Tp2X8Q&_nc_ss=73289&oh=00_Af3vztvsXZtNtWUr92vMyNkoKZTmh-rQFY6GPUqFwudUXg&oe=6A08C327" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
</channel>
</rss>http://localhost:1200/meta/ai/global-search/content_types=person - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — content_types=person</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/content_types=person" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/ (sort: RELEVANCE, total hits: 226). - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:09:51 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>Aaron Defazio</title>
<description>Aaron's research focuses on improving the practice of machine learning through the development of more reliable and theoretically sound methods such as ...</description>
<link>https://ai.meta.com/people/1115638629589333/aaron-defazio/</link>
<guid isPermaLink="false">1115638629589333</guid>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.35477-6/431543048_2088159024878120_1553708304445833368_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=109&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=vd7kajaxkn4Q7kNvwGQ7l84&_nc_oc=AdrVzy9eg4TqCHzxcCZ1L4XI2I740RBa3s9PgGhl0KVDLXIe-YZTwbQUpKKl868Ecec&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af3A6aqr00Dmv1_got2WgtZeuOk4nPbfXfw7k1f7kz-s_w&oe=69F449C8" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Speech & Audio</category>
<category>NLP</category>
<category>Theory</category>
<category>Core Machine Learning</category>
</item>
<item>
<title>Abhimanyu Dubey</title>
<description>Abhimanyu is a Research Scientist at Facebook AI Research working on problems in machine learning. He received his Ph.D. in 2021 from the Massachusetts ...</description>
<link>https://ai.meta.com/people/1401136157273652/abhimanyu-dubey/</link>
<guid isPermaLink="false">1401136157273652</guid>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.35477-6/433945609_1344738226210370_2541979406521528116_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=54biatSobnEQ7kNvwHm-noQ&_nc_oc=AdqgN3XWPgzN9zeEi07MRfAbFtwvMz1HwIPWMkl_i0ZncDiOKPKCljvu2nsTH9e6brE&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af1WwrzKnkdB19cnTQp2RXQ8J5rvTH_8vwI-2wCyP4Ib5Q&oe=69F4610D" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Computer Vision</category>
<category>Human & Machine Intelligence</category>
<category>Core Machine Learning</category>
<category>Research</category>
<category>ML Applications</category>
</item>
<item>
<title>Abhishek Das</title>
<description>Abhishek is a Research Scientist at Facebook AI Research (FAIR). His research focuses on deep learning and its applications in climate change, and in bu...</description>
<link>https://ai.meta.com/people/401335699167238/abhishek-das/</link>
<guid isPermaLink="false">401335699167238</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431537252_1543342826243118_6010155505132587328_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=102&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=7bKy_GgNvJUQ7kNvwHnc-io&_nc_oc=AdqcGdwhDx2dXwHqsEc4_GeDLVyC-LnWdC1bV1-bWeL-N8jcQe-nKtJ30FPvIdT16SM&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af1TKg_Gj1BZXPBB9aYlu7S-q6t7KkDsg5BaKnTAoGUQbw&oe=69F44FDB" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
<category>Research</category>
</item>
<item>
<title>Adam Polyak</title>
<description>Adam is a research engineer at Facebook AI Research (FAIR), focusing on deep learning, cross domain image generation, audio generation, and voice conver...</description>
<link>https://ai.meta.com/people/1434021433901222/adam-polyak/</link>
<guid isPermaLink="false">1434021433901222</guid>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.35477-6/431518452_1124340545357794_1947310908456905622_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=111&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=qnsH_8koyHsQ7kNvwFb8I9k&_nc_oc=AdqF4anURrXM4aReFHnW4Car9JsZOksY4Hr5eqp9BLYAY_vkyIocVQKUTHBgQv8u84c&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af1tOWdHxxE3g9L6gI8Zjx50P4iH4eTO5lglL1WHq3Gk9A&oe=69F4377A" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Research</category>
<category>Speech & Audio</category>
<category>ML Applications</category>
</item>
<item>
<title>Adeel Cheema</title>
<description>Adeel is a software engineer working on transparency &amp; control within Meta’s Responsible AI pillar. He focuses on creating product experiences to expl...</description>
<link>https://ai.meta.com/people/1808244339639715/adeel-cheema/</link>
<guid isPermaLink="false">1808244339639715</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431551867_767017845524520_8549663744281985032_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=3dHbSn4toEQQ7kNvwG-hxyL&_nc_oc=AdrneKtcRtDVD2x9eU9c0e9lX-OK623WzkPRI08JRJ3HANhfIiPoSRx5_FjSLcFEWTg&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af1kLwllLSM713z13uxZmgisMIeWvfbmFrG9NjpH-5KRbA&oe=69F44D79" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Adina Williams</title>
<description>Adina is a Research Scientist at Facebook AI Research in NYC (started October 2018). Previously, she earned her PhD at New York University in the Depart...</description>
<link>https://ai.meta.com/people/1396973444287406/adina-williams/</link>
<guid isPermaLink="false">1396973444287406</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431521840_2635057386653799_4484141047373535802_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=104&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=D9tP58vAF5kQ7kNvwHvASYA&_nc_oc=AdrKJZSIAfBDaQnc-wW1ToZsOSEhxJpnizTtfwAZJHHOlkTEt1y7LNyQXm2jzG7quUg&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af1dc3eXfWWhWpSSH9AFn9vMnGRKTom4j6t75YYUiReYtw&oe=69F467EC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Computer Vision</category>
<category>Core Machine Learning</category>
<category>Open Source</category>
<category>NLP</category>
<category>Conversational AI</category>
<category>Human & Machine Intelligence</category>
<category>ML Applications</category>
</item>
<item>
<title>Adriana Romero Soriano</title>
<description>Adriana is currently a research scientist at Facebook AI Research and an adjunct professor at McGill University. The goal of her research is to develop ...</description>
<link>https://ai.meta.com/people/425710613281765/adriana-romero-soriano/</link>
<guid isPermaLink="false">425710613281765</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431577893_1361257101250742_6288541672250118710_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=nsGSoSWY8rkQ7kNvwFMCuJe&_nc_oc=AdpfuVWkMsXwut0OSpmRwcFwPnozYpICadHJKwLA54PhHlVuRaKtTNxBtL3TcbLP6hk&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af0alPooyzDACIa-eYqOBZCmmBueuNZ-vXPct3-v4t9l7w&oe=69F44DCC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Core Machine Learning</category>
<category>Reinforcement Learni9ng</category>
<category>Conversational AI</category>
</item>
<item>
<title>Akhil Mathur</title>
<description>Akhil Mathur is a Research Scientist in the Llama Research team at Meta, working on enhancing the reasoning abilities of Llama models. Previously, as a ...</description>
<link>https://ai.meta.com/people/1251205439201018/akhil-mathur/</link>
<guid isPermaLink="false">1251205439201018</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/457008091_844600247808930_1170475198447093629_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=A2xSui7yUHAQ7kNvwHWOVVV&_nc_oc=Adp4eRq3NJ0287FNM_lETOYqZPQRg2pdbU9cuGT_PdgE67VDPw_dwbYD26U2CO3__R8&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af08heoInpvjK9yduIQaGyVH92UD6EghQgt-hBoIo_Bwgg&oe=69F4499C" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Akshara Rai</title>
<description>Akshara is a Research Scientist at Facebook AI Research, working at the intersection of machine learning and control. Her research aims at teaching robo...</description>
<link>https://ai.meta.com/people/423502423531745/akshara-rai/</link>
<guid isPermaLink="false">423502423531745</guid>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.35477-6/433923962_1381208202581063_6502596142308749692_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=0Wm1dG_MC68Q7kNvwGTCKdY&_nc_oc=AdoD6CrJ1m76HwlDd53FFb_r2h87MuGA8lBXoe2rT1bAb0Kfh-5LqBXoWovSqruL6lU&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af2iLOh_f1T-MoZyBAK-bbj8a-zxAm67mBhIfun0myLlyA&oe=69F43F4F" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Robotics</category>
<category>Computer Vision</category>
<category>Research</category>
<category>Reinforcement Learni9ng</category>
</item>
<item>
<title>Alborz Geramifard</title>
<description>Alborz Geramifard is a senior research manager at Meta AI supporting the Conversational AI. Prior to joining Meta, he led the conversational AI team at ...</description>
<link>https://ai.meta.com/people/363544420007153/alborz-geramifard/</link>
<guid isPermaLink="false">363544420007153</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431495520_3457483597876936_681607123433531609_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=QzBw61zVxHEQ7kNvwHFGECJ&_nc_oc=Adp_dD05XDi2euaYM6c0WcWx7YkqPn1LQ4QiGQnKLMbD88ygYqi7HyV0szo-h1D2zqM&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af3DzH9i8cEJIl3j39BREIM9mbSVi8T19jpAHs6Hyn6Uog&oe=69F45A51" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Conversational AI</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
</item>
<item>
<title>Alexander H. Miller</title>
<description>Alexander has worked in multi-domain research engineering management, supporting research engineers in London, Tel Aviv, and New York. He is advanced in...</description>
<link>https://ai.meta.com/people/264624513377272/alexander-h-miller/</link>
<guid isPermaLink="false">264624513377272</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/433950826_944833516851120_841662693953017766_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=6ftFVCshb-8Q7kNvwFmW3Vx&_nc_oc=AdpYZTkVT_RFDrpt-_sscuIcqOYLysabf-d4BJfV3EnXsHNOYkP_SRmSHFKxebGWjaU&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af1B69E3K8zkXfV05juIWjs8uD7lIdV1rhI_l57qZnHPfg&oe=69F45C6C" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Reinforcement Learni9ng</category>
<category>Graphics</category>
<category>NLP</category>
</item>
<item>
<title>Amey Porobo Dharwadker</title>
<description>Amey Porobo Dharwadker works as a Machine Learning Engineering Manager, leading Facebook's Video Recommendations Ranking team. Renowned for his expertis...</description>
<link>https://ai.meta.com/people/393924029921120/amey-porobo-dharwadker/</link>
<guid isPermaLink="false">393924029921120</guid>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.35477-6/431510720_7490594577724788_4289788014840147933_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=106&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=CDFvi0iDZRcQ7kNvwFhizrj&_nc_oc=Adr9sJsLUd39ruFsTcvrZXX1i_tYXLf-EyYjs5A_aQ4ACUVT9oNSSPcZpMVShrui3pk&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=XGv5Iwyzm4ZyemD-zediiA&_nc_ss=73289&oh=00_Af3WgwH4Qb0bzOJXyMML7ynhE6OO2qAzRvXru9QEDb4XYQ&oe=69F440F5" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
</channel>
</rss>... |
Contributor
http://localhost:1200/meta/ai/global-search/content_types=publication - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — content_types=publication</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/content_types=publication" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/ (sort: RELEVANCE, total hits: 1956). - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:09:56 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>AIRA₂: Overcoming Bottlenecks in AI Research Agents</title>
<description>Existing research has identified three structural performance bottlenecks in AI research agents: (1) synchronous single-GPU execution constrains sample throughput, limiting the benefit of search; (2) a generalization gap where validation-based selection causes overfitting and performance to degrade over extended search horizons; and (3) the limited capability of fixed, single-turn LLM operators imposes a ceiling on search performance. We introduce AIRA₂, which addresses these bottlenecks through three architectural choices: an asynchronous multi-GPU worker pool that increases experiment throughput linearly; a Hidden Consistent Evaluation protocol that delivers a reliable evaluation signal; and ReAct agents that dynamically scope their actions and debug interactively. On MLE-bench-30, AIRA₂ achieves a mean Percentile Rank of 81.5% at 24 hours and 83.1% at 72 hours, outperforming the strongest baseline, which achieves 72.7%. On AIRS-Bench, AIRA₂ exceeds human state-of-the-art on 6 out of 20 diverse research tasks. Ablations confirm that each architectural component is necessary, that performance follows a predictable scaling law that transfers across LLM backbones, and that the "overfitting" reported in prior work was driven by evaluation noise rather than true data memorization.</description>
<link>https://ai.meta.com/research/publications/aira-overcoming-bottlenecks-in-ai-research-agents/</link>
<guid isPermaLink="false">1298625375513718</guid>
<pubDate>Thu, 16 Apr 2026 07:00:00 GMT</pubDate>
<author>Karen Hambardzumyan, Nicolas Baldwin, Edan Toledo, Rishi Hazra, Michael Kuchnik, Bassel Al Omari, Thomas Simon Foster, Anton Protopopov, Jean-Christophe Gagnon-Audet, Ishita Mediratta, Kelvin Niu, Michael Shvartsman, Alisia Lupidi, Alexis Audran-Reiss, Parth Pathak, Tatiana Shavrina, Despoina Magka, Hela Momand, Derek Dunfield, Nicola Cancedda, Pontus Stenetorp, Carole-Jean Wu, Jakob Foerster, Yoram Bachrach, Martin Josifoski</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
</item>
<item>
<title>TransText: Transparency Aware Image-to-Video Typography Animation</title>
<description>We introduce the first method, to the best of our knowledge, for adapting image-to-video models to layer-aware text (glyph) animation, a capability critical for practical dynamic visual design. Existing approaches predominantly handle the transparency-encoding (alpha channel) as an extra latent dimension appended to the RGB space, necessitating the reconstruction of the underlying RGB-centric variational autoencoder (VAE). However, given the scarcity of high-quality transparent glyph data, retraining the VAE is computationally expensive and may erode the robust semantic priors learned from massive RGB corpora, potentially leading to latent pattern mixing. To mitigate these limitations, we propose TransText, a framework based on a novel Alpha-as-RGB paradigm to jointly model appearance and transparency without modifying the pre-trained generative manifold. TransText embeds the alpha channel as an RGB-compatible visual signal through latent spatial concatenation, explicitly ensuring strict cross-modal (RGB-and-Alpha) consistency while preventing feature entanglement. Our experiments demonstrate that TransText significantly outperforms baselines, generating coherent, high-fidelity transparent animations with diverse, fine-grained effects.</description>
<link>https://ai.meta.com/research/publications/transtext-transparency-aware-image-to-video-typography-animation/</link>
<guid isPermaLink="false">2172593836888174</guid>
<pubDate>Tue, 14 Apr 2026 07:00:00 GMT</pubDate>
<author>Fei Zhang, Zijian Zhou, Bohao Tang, Sen He, Hang Li (BizAI), Zhe Wang, Soubhik Sanyal, Pengfei Liu, Viktar Atliha, Tao Xiang, Frost Xu, Semih Gunel</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Computer Vision</category>
<category>ML Applications</category>
</item>
<item>
<title>Think in Strokes, Not Pixels: Process-Driven Image Generation via Interleaved Reasoning</title>
<description>Humans paint images incrementally: they plan a global layout, sketch a coarse draft, inspect, and refine details, and most importantly, each step is grounded in the evolving visual states. However, can unified multimodal models trained on text-image interleaved datasets also imagine the chain of intermediate states? In this paper, we introduce process-driven image generation, a multi-step paradigm that decomposes synthesis into an interleaved reasoning trajectory of thoughts and actions. Rather than generating images in a single step, our approach unfolds across multiple iterations, each consisting of 4 stages: textual planning, visual drafting, textual reflection, and visual refinement. The textual reasoning explicitly conditions how the visual state should evolve, while the generated visual intermediate in turn constrains and grounds the next round of textual reasoning. A core challenge of process-driven generation stems from the ambiguity of intermediate states: how can models evaluate each partially-complete image? We address this through dense, step-wise supervision that maintains two complementary constraints: for the visual intermediate states, we enforce the spatial and semantic consistency; for the textual intermediate states, we preserve the prior visual knowledge while enabling the model to identify and correct prompt-violating elements. This makes the generation process explicit, interpretable, and directly supervisable. To validate proposed method, we conduct experiments under various text-to-image generation benchmarks.</description>
<link>https://ai.meta.com/research/publications/think-in-strokes-not-pixels-process-driven-image-generation-via-interleaved-reasoning/</link>
<guid isPermaLink="false">847030005080644</guid>
<pubDate>Thu, 09 Apr 2026 07:00:00 GMT</pubDate>
<author>Lei Zhang, Junjiao Tian, Zhipeng Fan, Kunpeng Li, Jialiang Wang, Weifeng Chen, Markos Georgopoulos, Felix Xu, Yuxiao Bao, Julian McAuley, Manling Li, Zecheng He</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Computer Vision</category>
</item>
<item>
<title>A foundation model of vision, audition, and language for in-silico neuroscience</title>
<description>Cognitive neuroscience is fragmented into specialized models, each tailored to specific experimental paradigms, hence preventing a unified model of cognition in the human brain. Here, we introduce TRIBE v2, a tri-modal (video, audio and language) foundation model capable of predicting human brain activity in a variety of naturalistic and experimental conditions. Leveraging a unified dataset of over 1,000 hours of fMRI across 720 subjects, we demonstrate that our model accurately predicts high-resolution brain responses for novel stimuli, tasks and subjects, superseding traditional linear encoding models, delivering several-fold improvements in accuracy. Critically, TRIBE v2 enables in silico experimentation: tested on seminal visual and neuro-linguistic paradigms, it recovers a variety of results established by decades of empirical research. Finally, by extracting interpretable latent features, TRIBE v2 reveals the fine-grained topography of multisensory integration. These results establish artificial intelligence as a unifying framework for exploring the functional organization of the human brain.</description>
<link>https://ai.meta.com/research/publications/a-foundation-model-of-vision-audition-and-language-for-in-silico-neuroscience/</link>
<guid isPermaLink="false">1009839085553251</guid>
<pubDate>Thu, 26 Mar 2026 07:00:00 GMT</pubDate>
<author>Stéphane d'Ascoli, Jérémy Rapin, Yohann Benchetrit, Teon Brooks, Katelyn Begany, Josephine Raugel, Hubert Jacob Banville, Jean Remi King</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
</item>
<item>
<title>HyperAgents</title>
<description>Self-improving AI systems aim to reduce reliance on human engineering by learning to improve their own learning and problem-solving processes. Existing approaches to recursive self-improvement typically rely on fixed, handcrafted meta-level mechanisms, which fundamentally limit how fast such systems can improve. The Darwin Gödel Machine (DGM)(Zhang et al., 2025b) demonstrates that open-ended self-improvement is achievable in coding. Starting from a single coding agent, the DGM repeatedly generates and evaluates self-modified variants, forming a growing archive of stepping stones for future improvement. Because both evaluation and self-modification are coding tasks, gains in coding ability can translate into gains in self-improvement ability. However, this alignment does not generally hold beyond coding domains.
We introduce hyperagents, self-referential agents that integrate a task agent (which solves the target task) and a meta agent (which modifies itself and the task agent) into a single editable program. Crucially, the meta-level modification procedure is itself editable, enabling metacognitive self-modification, improving not only task-solving behavior, but also the mechanism that generates future improvements. We instantiate this framework by extending DGM to create DGM-Hyperagents (DGM-H). By allowing the improvement procedure to evolve, the DGM-H eliminates the assumption of domain-specific alignment between task performance and self-modification skill, and can potentially support self-accelerating progress on any computable task. Across diverse domains (coding, paper review, robotics reward design, and Olympiad-level math-solution grading), the DGM-H improves performance over time and outperforms baselines without self-improvement or open-ended exploration, as well as prior self-improving systems like DGM. We further show that the DGM-H improves the process by which it generates new agents (e.g., persistent memory, performance tracking), and that these meta-level improvements transfer across domains and accumulate across runs. All experiments were conducted with safety precautions (e.g., sandboxing, human oversight). We discuss what safety entails in this setting and the broader implications of self-improving systems. DGM-Hyperagents offer a glimpse of open-ended AI systems that do not merely search for better solutions, but continually improve their search for how to improve.</description>
<link>https://ai.meta.com/research/publications/hyperagents/</link>
<guid isPermaLink="false">1669124537439877</guid>
<pubDate>Tue, 24 Mar 2026 07:00:00 GMT</pubDate>
<author>Jenny Zhang, Bingchen Zhao, Winnie Yang, Jakob Foerster, Sam Devlin, Tatiana Shavrina</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
<category>Open Source</category>
</item>
<item>
<title>Omnilingual MT: Machine Translation for 1,600 Languages</title>
<description>Advances made through No Language Left Behind (NLLB) have demonstrated that high-quality machine translation (MT) scale to 200 languages. Later Large Language Models (LLMs) have been adopted for MT, increasing in quality but not necessarily extending language coverage. Current systems remain constrained by limited coverage and a persistent generation bottleneck: while crosslingual transfer enables models to somehow understand many undersupported languages, they often cannot generate them reliably, leaving most of the world’s 7,000 languages—especially endangered and marginalized ones—outside the reach of modern MT. Early explorations in extreme scaling offered promising proofs of concept but did not yield sustained solutions.
We present Omnilingual Machine Translation (OMT), the first MT system supporting more than 1,600 languages. This scale is enabled by a comprehensive data strategy that integrates large public multilingual corpora with newly created datasets, including manually curated MeDLEY bitext, synthetic backtranslation, and mining, substantially expanding coverage across long-tail languages, domains, and registers. To ensure both reliable and expansive evaluation, we combined standard metrics with a suite of evaluation artifacts: BLASER 3 quality estimation model (reference-free), OmniTOX toxicity classifier, BOUQuET dataset (a newly created, largest-to-date multilingual evaluation collection built from scratch and manually extended across a wide range of linguistic families), and Met-BOUQuET dataset (faithful multilingual quality estimation at scale). We explore two ways of specializing an LLM for machine translation: as a decoder-only model (OMT-LLaMA) or as a module in an encoder–decoder architecture (OMT-NLLB). The former consists of a model built on LLaMA3, with multilingual continual pretraining and retrieval-augmented translation for inference-time adaptation. The latter is a model built on top of a multilingual aligned space (OmniSONAR, itself also based on LLaMA3), and introduces a training methodology that can exploit non-parallel data, allowing us to incorporate the decoder-only continuous pretraining data into the training of an encoder–decoder architecture. Notably, all our 1B to 8B parameter models match or exceed the MT performance of a 70B LLM baseline, revealing a clear specialization advantage and enabling strong translation quality in low-compute settings. Moreover, our evaluation of English-to-1,600 translations further shows that while baseline models can interpret undersupported languages, they frequently fail to generate them with meaningful fidelity; OMT-LLaMA models substantially expand the set of languages for which coherent generation is feasible. Additionally, OMT models improve in cross-lingual transfer, being close to solving the “understanding” part of the puzzle in MT for the 1,600 evaluated. Beyond strong out-of-the-box performance, we find that finetuning and retrieval-augmented generation offer additional pathways to improve quality for the given subset of languages when targeted data or domain knowledge is available. Our leaderboard and main humanly created evaluation datasets (BOUQuET and Met-BOUQuET) are dynamically evolving towards Omnilinguality and freely available.</description>
<link>https://ai.meta.com/research/publications/omnilingual-mt-machine-translation-for-1600-languages/</link>
<guid isPermaLink="false">1449924233245450</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual MT Team, Belen Alastruey, Niyati Bafna, Andrea Caciolai, Kevin Heffernan, Artyom Kozhevnikov, Christophe Ropers, Eduardo Sánchez, Charles-Eric Saint-James, Ioannis Tsiamas, Chierh CHENG, Joe Chuang, Paul-Ambroise Duquenne, Mark Duppenthaler, Nate Ekberg, Cynthia Gao, Pere Lluís Huguet Cabot, João Maria Janeiro, Jean Maillard, Gabriel Mejia Gonzalez, Holger Schwenk, Edan Toledo, Arina Turkatenko, Albert Ventayol-Boada, Rashel Moritz, Alexandre Mourachko, Surya Parimi, Mary Williamson, Shireen Yates, David Dale, Marta R. Costa-jussa</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Omnilingual SONAR: Cross-Lingual and Cross-Modal Sentence Embeddings Bridging Massively Multilingual Text and Speech</title>
<description>Cross-lingual sentence encoders have traditionally been limited to a few hundred languages, and have sacrificed downstream performance to achieve better alignment across languages, limiting their adoption. In this work, we introduce OmniSONAR, a novel family of omnilingual, cross-lingual and cross-modal sentence embedding models that breaks this barrier. We establish a unified semantic space, natively encompassing text, speech, code and mathematical expressions, while achieving state-of-the-art downstream performance for an unprecedented scale of thousands of languages, from high-resource languages to extremely low-resource varieties.
To achieve this scale without representation collapse and while maintaining top-tier performance in the high-resource languages, we employ a progressive training strategy. We first build a state-of-the-art foundational embedding space for 200 languages using an LLM-initialized Encoder-Decoder, combining token-level decoding with a novel split-softmax contrastive loss and synthetic hard negatives. Leveraging this strong foundational space, we expand to several thousands of language varieties via a specialized two-stage teacher-student encoder distillation framework. Further modeling extensions derived from OmniSONAR address long context inputs and token-centric representations. Finally, we demonstrate the cross-modal extensibility of this space by seamlessly mapping 177 spoken languages into it.
OmniSONAR redefines the state of the art for multilingual representation learning. It halves the cross-lingual similarity search error rate of the previous best models on the 200 languages of FLORES, while also achieving a staggering 15-fold error rate reduction across 1,560 languages in the BIBLE benchmark. Furthermore, our embedding model enables unprecedented translation capabilities, outperforming NLLB-3B on several multilingual benchmarks, and surpassing all previous models, including multi-billion-parameter LLMs, by 15 chrF++ points in 1,560→English translation in the BIBLE benchmark. Beyond alignment and translation, OmniSONAR demonstrates strong general-purpose capabilities across downstream embedding tasks on MTEB and programming languages on XLCoST. For the speech modality, our massively multilingual extension exhibits a 43% lower error rate in cross-lingual and cross-modal similarity search, while achieving 97% of SeamlessM4T performance in speech-to-text translation, despite being a zero-shot translation model trained only with ASR data. Finally, by training an encoder-decoder language model, Spectrum, exclusively on English text that processes OmniSONAR sequences, we unlock immediate high-performance transfer to thousands of languages and the speech modality for complex downstream tasks. These outstanding results position OmniSONAR as a robust, language- and modality-agnostic foundation for any downstream usage.</description>
<link>https://ai.meta.com/research/publications/omnilingual-sonar-cross-lingual-and-cross-modal-sentence-embeddings-bridging-massively-multilingual-text-and-speech/</link>
<guid isPermaLink="false">1231952259116893</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual SONAR Team, João Maria Janeiro, Pere Lluís Huguet Cabot, Ioannis Tsiamas, Yen Meng, Vivek Iyer, Guillem Ramirez, Loic Barrault, Belen Alastruey, Yu-An Chung, Marta R. Costa-jussa, David Dale, Kevin Heffernan, Jaehyeong Jo, Artyom Kozhevnikov, Alexandre Mourachko, Christophe Ropers, Holger Schwenk, Paul-Ambroise Duquenne</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Speech & Audio</category>
</item>
<item>
<title>Unified Vision–Language Modeling via Concept Space Alignment</title>
<description>We introduce v-Sonar, a vision–language embedding space extended from the text-only embedding space Sonar (Omnilingual Embeddings Team et al., 2026), which supports 1500 text languages and 177 speech languages. To construct v-Sonar, we propose a post-hoc alignment pipeline that maps the representations of an existing vision encoder into the Sonar space. We thoroughly evaluate v-Sonar and show that its embeddings achieve competitive performance on text-to-video retrieval. Equipped with the Sonar text decoder, v-Sonar further surpasses state-of-the-art vision–language models on video captioning tasks, including Dream-1k (Bleu 24.3 vs. 19.6) and Vatex (Bleu 45.0 vs. 41.5).
Leveraging v-Sonar, we first demonstrate that the Large Concept Model (LCM; LCM team et al. 2024) operating in Sonar and trained with English text only, can perform both single- and multi-visual concept understanding in a zero-shot manner. Finally, we introduce v-LCM, which extends the LCM with vision–language instruction tuning. v-LCM encodes vision and language inputs into an unified sequence of latent embeddings via v-Sonar and Sonar, and it is trained with the same latent diffusion objective for next-embedding prediction as in LCM’s text-only pre-training. Experiments on a large-scale multilingual and -modal instruction–tuning data mixture highlight the potential of v-LCM: v-LCM matches state-of-the-art vision-language models on tasks covering image/video captioning and question answering, while significantly outperforming them across 61 rich- to low-resource languages out of all 62 tested languages.</description>
<link>https://ai.meta.com/research/publications/unified-vision-language-modeling-via-concept-space-alignment/</link>
<guid isPermaLink="false">1257213679668978</guid>
<pubDate>Fri, 27 Feb 2026 08:00:00 GMT</pubDate>
<author>Yifu Qiu, Paul-Ambroise Duquenne, Holger Schwenk</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Research</category>
</item>
<item>
<title>Learning Personalized Agents from Human Feedback</title>
<description>Modern AI agents are powerful but often fail to align with the idiosyncratic, evolving preferences of individual users. Prior approaches typically rely on static datasets, either training implicit preference models on interaction history or encoding user profiles in external memory. However, these approaches struggle with new users and with preferences that change over time. We introduce Personalized Agents from Human Feedback (PAHF), a framework for continual personalization in which agents learn online from live interaction using explicit per-user memory. PAHF operationalizes a three-step loop: (1) seeking pre-action clarification to resolve ambiguity, (2) grounding actions in preferences retrieved from memory, and (3) integrating post-action feedback to update memory when preferences drift. To evaluate this capability, we develop a four-phase protocol and two benchmarks in embodied manipulation and online shopping. These benchmarks quantify an agent's ability to learn initial preferences from scratch and subsequently adapt to persona shifts. Our theoretical analysis and empirical results show that integrating explicit memory with dual feedback channels is critical: PAHF learns substantially faster and consistently outperforms both no-memory and single-channel baselines, reducing initial personalization error and enabling rapid adaptation to preference shifts.</description>
<link>https://ai.meta.com/research/publications/learning-personalized-agents-from-human-feedback/</link>
<guid isPermaLink="false">910245944695813</guid>
<pubDate>Thu, 26 Feb 2026 08:00:00 GMT</pubDate>
<author>Kaiqu Liang, Julia Kruk, Shengyi Qian, Xianjun Yang, Shengjie Bi, Shaoliang Nie, Michael Zhang, Lijuan Liu, Jaime Fernández Fisac, Shuyan Zhou, Saghar Hosseini</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Conversational AI</category>
<category>Research</category>
</item>
<item>
<title>FERRET: Framework for Expansion Reliant Red Teaming</title>
<description>We introduce a multi-faceted automated red teaming framework in which the goal is to generate multi-modal adversarial conversations that would break a target model and introduce various expansions that would result in more effective and efficient adversarial conversations. The introduced expansions include: 1. Horizontal expansion in which the goal is for the red team model to self-improve and generate more effective conversation starters that would shape a conversation. 2. Vertical expansion in which the goal is to take these conversation starters that are discovered in the horizontal expansion phase and expand them into effective multi-modal conversations and 3. Meta expansion in which the goal is for the red team model to discover more effective multi-modal attack strategies during the course of a conversation. We call our framework FERRET (Framework for Expansion Reliant Red Teaming) and compare it with various existing automated red teaming approaches. In our experiments, we demonstrate the effectiveness of FERRET in generating effective multi-modal adversarial conversations and its superior performance against existing state of the art approaches.</description>
<link>https://ai.meta.com/research/publications/ferret-framework-for-expansion-reliant-red-teaming/</link>
<guid isPermaLink="false">2396003004185464</guid>
<pubDate>Fri, 13 Feb 2026 08:00:00 GMT</pubDate>
<author>Ninareh Mehrabi, Vítor Albiero, Maya Pavlova, Joanna Bitton</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Responsible AI</category>
</item>
<item>
<title>UniT: Unified Multimodal Chain-of-Thought Test-time Scaling</title>
<description>Unified models can handle both multimodal understanding and generation within a single architecture, yet they typically operate in a single pass without iteratively refining their outputs. Many multimodal tasks, especially those involving complex spatial compositions, multiple interacting objects, or evolving instructions, require decomposing instructions, verifying intermediate results, and making iterative corrections. While test-time scaling (TTS) has demonstrated that allocating additional inference compute for iterative reasoning substantially improves language model performance, extending this paradigm to unified multimodal models remains an open challenge.
We introduce UniT, a framework for multimodal chain-of-thought test-time scaling that enables a single unified model to reason, verify, and refine across multiple rounds. UniT combines agentic data synthesis, unified model training, and flexible test-time inference to elicit cognitive behaviors including verification, subgoal decomposition, and content memory. Our key findings are: (1) unified models trained on short reasoning trajectories generalize to longer inference chains at test time; (2) sequential chain-of-thought reasoning provides a more scalable and compute-efficient TTS strategy than parallel sampling; (3) training on generation and editing trajectories improves out-of-distribution visual reasoning. These results establish multimodal test-time scaling as an effective paradigm for advancing both generation and understanding in unified models.</description>
<link>https://ai.meta.com/research/publications/unit-unified-multimodal-chain-of-thought-test-time-scaling/</link>
<guid isPermaLink="false">1477019817147720</guid>
<pubDate>Wed, 11 Feb 2026 08:00:00 GMT</pubDate>
<author>Leon Liangyu Chen, Haoyu Ma, Zhipeng Fan, Ziqi Huang, Animesh Sinha, Xiaoliang Dai, Jialiang Wang, Zecheng He, Jianwei Yang, Chunyuan Li, Junzhe Sun, Chu Wang, Serena Yeung-Levy, Felix Juefei-Xu</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Computer Vision</category>
</item>
<item>
<title>AIRS-Bench: a Suite of Tasks for Frontier AI Research Science Agents</title>
<description>LLM agents hold significant promise for advancing scientific research. To accelerate this progress, we introduce AIRS-Bench (the AI Research Science Benchmark), a suite of 20 tasks sourced from state-of-the-art machine learning papers. These tasks span diverse domains, including language modeling, mathematics, bioinformatics, and time series forecasting. AIRS-Bench tasks assess agentic capabilities over the full research lifecycle---including idea generation, experiment analysis and iterative refinement---without providing baseline code. The AIRS-Bench task format is versatile, enabling easy integration of new tasks and rigorous comparison across different agentic frameworks. We establish baselines using frontier models paired with both sequential and parallel scaffolds. Our results show that agents exceed human SOTA in four tasks but fail to match it in sixteen others. Even when agents surpass human benchmarks, they do not reach the theoretical performance ceiling for the underlying tasks. These findings indicate that AIRS-Bench is far from saturated and offers substantial room for improvement.
We open-source the AIRS-Bench task definitions and evaluation code to catalyze further development in autonomous scientific research.</description>
<link>https://ai.meta.com/research/publications/airs-bench-a-suite-of-tasks-for-frontier-ai-research-science-agents/</link>
<guid isPermaLink="false">2685677328474642</guid>
<pubDate>Tue, 10 Feb 2026 08:00:00 GMT</pubDate>
<author>Alisia Lupidi, Bhavul Gauri, Thomas Simon Foster, Bassel Al Omari, Despoina Magka, Alberto Pepe, Alexis Audran-Reiss, Muna Aghamelu, Nicolas Baldwin, Lucia Cipolina-Kun, Jean-Christophe Gagnon-Audet, Chee Hau Leow, Sandra Lefdal, Hossam Mossalam, Abhinav Moudgil, Saba Nazir, Emanuel Tewolde, Isabel Urrego, Jordi Armengol-Estape, Amar Budhiraja, Gaurav Chaurasia, Abhishek Charnalia, Derek Dunfield, Karen Hambardzumyan, Daniel Izcovich, Martin Josifoski, Ishita Mediratta, Kelvin Niu, Parth Pathak, Michael Shvartsman, Edan Toledo, Anton Protopopov, Roberta Raileanu, Alexander Miller, Tatiana Shavrina, Jakob Foerster, Yoram Bachrach</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
</item>
<item>
<title>Energy-Based Models for Atomic-Resolution Protein Conformations | Facebook AI Research</title>
<description>We propose an energy-based model (EBM) of protein conformations that operates at atomic scale. The model is trained solely on crystallized protein data.…</description>
<link>https://ai.meta.com/research/publications/energy-based-models-for-atomic-resolution-protein-conformations/</link>
<guid isPermaLink="false">237760337463058</guid>
<pubDate>Sat, 25 Apr 2020 07:00:00 GMT</pubDate>
<author>Yilun Du, Joshua Meier, Jerry Ma, Rob Fergus, Alexander Rives</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
</item>
<item>
<title>Scaling up online speech recognition using ConvNets | Facebook AI Research</title>
<description>We design an online end-to-end speech recognition system based on Time-Depth Separable (TDS) convolutions and Connectionist Temporal Classification (CTC). The system has almost three times the throughput of a well tuned hybrid ASR baseline…</description>
<link>https://ai.meta.com/research/publications/scaling-up-online-speech-recognition-using-convnets/</link>
<guid isPermaLink="false">623727015029499</guid>
<pubDate>Mon, 13 Jan 2020 08:00:00 GMT</pubDate>
<author>Vineel Pratap, Qiantong Xu, Jacob Kahn, Gilad Avidov, Tatiana Likhomanenko, Awni Hannun, Vitaliy Liptchinsky, Gabriel Synnaeve, Ronan Collobert</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Facebook AI's WAT19 Myanmar-English Translation Task Submission</title>
<description>This paper describes Facebook AI's submission to the WAT 2019 Myanmar-English translation task. Our baseline systems are BPE-based transformer models.…</description>
<link>https://ai.meta.com/research/publications/facebook-ai-wat19-myanmar-english-translation-task-submission/</link>
<guid isPermaLink="false">389591045439111</guid>
<pubDate>Thu, 31 Oct 2019 07:00:00 GMT</pubDate>
<author>Peng-Jen Chen, Jiajun Shen, Matt Le, Vishrav Chaudhary, Ahmed El-Kishky, Guillaume Wenzek, Myle Ott, Marc’Aurelio Ranzato</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Order-Aware Generative Modeling Using the 3D-Craft Dataset | Facebook AI Research</title>
<description>Research on 2D and 3D generative models typically focuses on the final artifact being created, e.g., an image or a 3D structure. Unlike 2D image generation, the generation of 3D objects in the real world is commonly constrained by the process…</description>
<link>https://ai.meta.com/research/publications/order-aware-generative-modeling-using-the-3d-craft-dataset/</link>
<guid isPermaLink="false">1247541865430840</guid>
<pubDate>Sun, 27 Oct 2019 07:00:00 GMT</pubDate>
<author>Zhuoyuan Chen, Demi Guo, Tong Xiao, Saining Xie, Xinlei Chen, Haonan Yu, Jonathan Gray, Kavya Srinet, Haoqi Fan, Jerry Ma, Charles R. Qi, Shubham Tulsiani, Arthur Szlam, Larry Zitnick</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
</item>
<item>
<title>ELF OpenGo: An Analysis and Open Reimplementation of AlphaZero | Facebook AI Research</title>
<description>The AlphaGo, AlphaGo Zero, and AlphaZero series of algorithms are remarkable demonstrations of deep reinforcement learning’s capabilities, achieving superhuman performance in the complex game of Go with progressively increasing autonomy.…</description>
<link>https://ai.meta.com/research/publications/elf-opengo-an-analysis-and-open-reimplementation-of-alpha-zero/</link>
<guid isPermaLink="false">2747507891982858</guid>
<pubDate>Tue, 11 Jun 2019 07:00:00 GMT</pubDate>
<author>Yuandong Tian, Jerry Ma, Qucheng Gong, Shubho Sengupta, Zhuoyuan Chen, James Pinkerton, Larry Zitnick</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Computer Vision</category>
</item>
<item>
<title>Equi-normalization of Neural Networks | Facebook AI Research</title>
<description>Modern neural networks are over-parametrized. In particular, each rectified linear hidden unit can be modified by a multiplicative factor by adjusting input and output weights, without changing the rest of the network. Inspired by the…</description>
<link>https://ai.meta.com/research/publications/equi-normalization-of-neural-networks/</link>
<guid isPermaLink="false">2618277071589238</guid>
<pubDate>Sun, 05 May 2019 07:00:00 GMT</pubDate>
<author>Pierre Stock, Benjamin Graham, Rémi Gribonval, Hervé Jégou</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
<category>Computer Vision</category>
</item>
<item>
<title>On the Pitfalls of Measuring Emergent Communication | Facebook AI Research</title>
<description>How do we know if communication is emerging in a multi-agent system? The vast majority of recent papers on emergent communication show that adding a communication channel leads to an increase in reward or task success. This is a useful…</description>
<link>https://ai.meta.com/research/publications/on-the-pitfalls-of-measuring-emergent-communication/</link>
<guid isPermaLink="false">2451938888356370</guid>
<pubDate>Thu, 14 Mar 2019 07:00:00 GMT</pubDate>
<author>Ryan Lowe, Jakob Foerster, Y-Lan Boureau, Joelle Pineau, Yann Dauphin</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Loss in Translation: Learning Bilingual Word Mapping with a Retrieval Criterion | Facebook AI Research</title>
<description>Continuous word representations learned separately on distinct languages can be aligned so that their words become comparable in a common space. Existing works typically solve a least-square regression problem to learn a rotation aligning a…</description>
<link>https://ai.meta.com/research/publications/loss-in-translation-learning-bilingual-word-mapping-with-a-retrieval-criterion/</link>
<guid isPermaLink="false">1432754183558579</guid>
<pubDate>Tue, 30 Oct 2018 07:00:00 GMT</pubDate>
<author>Armand Joulin, Piotr Bojanowski, Tomas Mikolov, Hervé Jégou</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>LAMV: Learning to align and match videos with kernelized temporal layers | Facebook AI Research</title>
<description>This paper considers a learnable approach for comparing and aligning videos. Our architecture builds upon and revisits temporal match kernels within neural networks: we propose a new temporal layer that finds temporal alignments by maximizing…</description>
<link>https://ai.meta.com/research/publications/lamv-learning-to-align-and-match-videos-with-kernelized-temporal-layers/</link>
<guid isPermaLink="false">2177100492586051</guid>
<pubDate>Tue, 19 Jun 2018 07:00:00 GMT</pubDate>
<author>Lorenzo Baraldi, Matthijs Douze, Rita Cucchiara, Hervé Jégou</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Computer Vision</category>
<category>Research</category>
</item>
<item>
<title>Low-shot learning with large-scale diffusion | Facebook AI Research</title>
<description>This paper considers the problem of inferring image labels from images when only a few annotated examples are available at training time. This setup is often referred to as low-shot learning, where a standard approach is to re-train the last…</description>
<link>https://ai.meta.com/research/publications/low-shot-learning-with-large-scale-diffusion/</link>
<guid isPermaLink="false">534360120449321</guid>
<pubDate>Mon, 18 Jun 2018 07:00:00 GMT</pubDate>
<author>Matthijs Douze, Arthur Szlam, Bharath Hariharan, Hervé Jégou</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Speech & Audio</category>
<category>Computer Vision</category>
<category>Research</category>
</item>
<item>
<title>Mastering the Dungeon: Grounded Language Learning by Mechanical Turker Descent | Facebook AI Research</title>
<description>Contrary to most natural language processing research, which makes use of static datasets, humans learn language interactively, grounded in an environment. In this work we propose an interactive learning procedure called Mechanical…</description>
<link>https://ai.meta.com/research/publications/mastering-the-dungeon-grounded-language-learning-by-mechanical-turker-descent/</link>
<guid isPermaLink="false">2704786416424529</guid>
<pubDate>Mon, 30 Apr 2018 07:00:00 GMT</pubDate>
<author>Zhilin Yang, Saizheng Zhang, Jack Urbanek, Will Feng, Alexander H. Miller, Arthur Szlam, Douwe Kiela, Jason Weston</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
<category>Computer Vision</category>
</item>
<item>
<title>Polysemous Codes | Facebook AI Research</title>
<description>This paper considers the problem of approximate nearest neighbor search in the compressed domain. We introduce polysemous codes, which offer both the distance estimation quality of product quantization and the efficient comparison of binary…</description>
<link>https://ai.meta.com/research/publications/polysemous-codes/</link>
<guid isPermaLink="false">617678119003343</guid>
<pubDate>Mon, 10 Oct 2016 07:00:00 GMT</pubDate>
<author>Matthijs Douze, Hervé Jégou, Florent Perronnin</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=gXCxukPpSa-q8seiN3pPQw&_nc_ss=73289&oh=00_Af06Gcj3zkmnGWfrKQjo-owEk_77btwRRSb4cSUapZre7A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Speech & Audio</category>
<category>Computer Vision</category>
<category>Research</category>
</item>
</channel>
</rss>http://localhost:1200/meta/ai/global-search/content_types=dataset - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — content_types=dataset</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/content_types=dataset" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/ (sort: RELEVANCE, total hits: 44). - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:10:00 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>Multi-room Apartments Simulation (MRAS) Dataset | Meta AI Research</title>
<description>The Multi-Room Apartments Simulation (MRAS) dataset is a multi-modal dataset created for the task of estimating spatially-distributed acoustic parameters in complex scenes.</description>
<link>https://ai.meta.com/datasets/mras-simulated-rooms-dataset/</link>
<guid isPermaLink="false">2247220155683256</guid>
<pubDate>Wed, 28 Jan 2026 18:49:18 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/612410014_2220471548441535_606775622330495537_n.jpg?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=1bE1l45d6ZQQ7kNvwGMtrFY&_nc_oc=Adqxxb2bGLw1oq3NG8aZYuMODgbpt__dmsQw5pkrsOQbRgeHEIStu0h09rsY2lXjCgg&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=BUcY_y2EZktQqJvfTCc1Jw&_nc_ss=73289&oh=00_Af3ZzgLlkJC1bYG1kwLkbAqHFbiC97Q1ByfYaMd9AS6ClQ&oe=6A08CB19" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
<item>
<title>Meta Synthetic Environments Lidar Dataset | Meta AI Research</title>
<description>The Meta Synthetic Environments (MSE) Lidar Dataset is the first-of-its-kind large-scale single-photon lidar dataset, built on top of Aria Synthetic Environments (ASE) and intended to unlock new machine learning capabilities for single-photon lidars.</description>
<link>https://ai.meta.com/datasets/meta-synthetic-environments-lidar-dataset/</link>
<guid isPermaLink="false">1350377946838590</guid>
<pubDate>Wed, 17 Dec 2025 17:38:04 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/579919244_838469262096799_1005035704573286504_n.jpg?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=o9JbB5ingBsQ7kNvwHA0V62&_nc_oc=AdqPrV3qsBWRkJ--X3NurSHlwQnFzOs4YHaT-h7h0R26hiLf0O0w3JR9O0m4pC02Emk&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=BUcY_y2EZktQqJvfTCc1Jw&_nc_ss=73289&oh=00_Af22j93x9LCKu9sKXPhGb11ktm7Gl3DqK-mavGoXDM6Nwg&oe=6A08C2C4" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
<item>
<title>PLM Data | Meta AI Research</title>
<description>PLM Data is a comprehensive collection of synthetic and human-annotated datasets for detailed visual understanding, combining existing and newly collected data for tasks like OCR, captioning, and visual question answering.</description>
<link>https://ai.meta.com/datasets/plm-data/</link>
<guid isPermaLink="false">672877175127332</guid>
<pubDate>Tue, 27 May 2025 16:18:53 GMT</pubDate>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.2365-6/384127954_1010009116964035_2906046058102555347_n.jpg?_nc_cat=101&ccb=1-7&_nc_sid=e280be&_nc_ohc=iT1pu9-tiuUQ7kNvwHtFA10&_nc_oc=Adr8RG80qqN300ZkIHjGMPyqhCsZCEL4KCtLpgmVJWBLBXiQOTlV3RRqhHsAzEtj54M&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=BUcY_y2EZktQqJvfTCc1Jw&_nc_ss=73289&oh=00_Af29vq6nf8CKyHVXqPfzPZRgqPgEqnPF9yNukyu_KFT7qg&oe=6A08DD05" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
<item>
<title>PE Video | Meta AI Research</title>
<description>The PE Video Dataset (PVD) is a large-scale collection of 1 million diverse videos, featuring 120,000+ expertly annotated clips. The dataset was introduced in our paper "Perception Encoder".</description>
<link>https://ai.meta.com/datasets/pe-video/</link>
<guid isPermaLink="false">29433953799551622</guid>
<pubDate>Thu, 17 Apr 2025 18:01:37 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/491815770_995150272810861_1471459038704351685_n.jpg?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=Yffaq2cg7KYQ7kNvwHOCq7_&_nc_oc=AdpzWwSSLkwiTp64jDXivlaSzHsaqN7C4GFiFG5vRvBgQA6DBbdI6qCgFcRb6naw5I0&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=BUcY_y2EZktQqJvfTCc1Jw&_nc_ss=73289&oh=00_Af3PM7-JyCv8OawESPjh7boqlPNgMttvzmZYaIYyEcG8xw&oe=6A08D0C8" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
</channel>
</rss>... |
Contributor
http://localhost:1200/meta/ai/global-search/content_types=tool - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — content_types=tool</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/content_types=tool" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/ (sort: RELEVANCE, total hits: 45). - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:10:03 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>FastText</title>
<description>A lightweight library designed to help build scalable solutions for text representation and classification.</description>
<link>https://ai.meta.com/tools/fasttext/</link>
<guid isPermaLink="false">842058939461581</guid>
<pubDate>Tue, 14 Oct 2025 23:29:35 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=OqfqypXIIwixdRENqBn2sg&_nc_ss=73289&oh=00_Af2_8kSB_zxFVSmfG69Wdg9FkXGf_V-jiMBeLNEx3e_6gQ&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>DensePose</title>
<description>DensePose, dense human pose estimation, is designed to map all human pixels of an RGB image to a 3D surface-based representation of the human body.</description>
<link>https://ai.meta.com/tools/densepose/</link>
<guid isPermaLink="false">368589657320353</guid>
<pubDate>Tue, 14 Oct 2025 22:00:22 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=OqfqypXIIwixdRENqBn2sg&_nc_ss=73289&oh=00_Af2_8kSB_zxFVSmfG69Wdg9FkXGf_V-jiMBeLNEx3e_6gQ&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>KILT Benchmarking</title>
<description>KILT is a resource for training, evaluating and analyzing NLP models on Knowledge Intensive Language Tasks.</description>
<link>https://ai.meta.com/tools/kilt/</link>
<guid isPermaLink="false">2743325952604548</guid>
<pubDate>Thu, 11 Sep 2025 21:42:53 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/119073509_368822947476971_6922122926535022194_n.png?_nc_cat=106&ccb=1-7&_nc_sid=e280be&_nc_ohc=Mg93n9KMtaEQ7kNvwFSR1DD&_nc_oc=Adpz_dkMOZt7XCaD3IY-NDACjzb_yjD9MqF9YeAQe4X8m1D7s3vDcnl8HpJng1FiwjI&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=OqfqypXIIwixdRENqBn2sg&_nc_ss=73289&oh=00_Af0hoUO0RQCWJlJhyTyQZvFhfR3qTq74g_8hX44kXVFgKw&oe=6A08CF54" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>Faiss</title>
<description>A library that allows developers to quickly search for embeddings of multimedia documents that are similar to each other.</description>
<link>https://ai.meta.com/tools/faiss/</link>
<guid isPermaLink="false">297198177617190</guid>
<pubDate>Wed, 05 Feb 2025 00:52:31 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=OqfqypXIIwixdRENqBn2sg&_nc_ss=73289&oh=00_Af2_8kSB_zxFVSmfG69Wdg9FkXGf_V-jiMBeLNEx3e_6gQ&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>System Cards - Meta AI</title>
<description>System cards or multiple machine learning models help users understand the intention, impact and limitations of our AI systems.</description>
<link>https://ai.meta.com/tools/system-cards/</link>
<guid isPermaLink="false">632380378064785</guid>
<pubDate>Tue, 12 Dec 2023 18:58:53 GMT</pubDate>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.2365-6/343413316_1234589910516703_2524736655913484825_n.jpg?_nc_cat=101&ccb=1-7&_nc_sid=e280be&_nc_ohc=IaCyxndRw5IQ7kNvwFj8rrw&_nc_oc=AdpZ1s_l8Qd5oK3LktrQP227GaJFpnXG93GoP1Y_Q84ZFuyDLHxbm9KymLRE4vp_V_M&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=OqfqypXIIwixdRENqBn2sg&_nc_ss=73289&oh=00_Af0NO1WzhZ868M5p7Q7BCI3alo3whrWIwQZzhwOCsKTCWQ&oe=6A08D680" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>Segment Anything Dataset</title>
<description>Segment Anything 1 Billion (SA-1B) is a dataset designed for training general-purpose object segmentation models from open world images. The dataset was introduced in our paper “Segment Anything”.</description>
<link>https://ai.meta.com/tools/segment-anything-dataset/</link>
<guid isPermaLink="false">1375631516311803</guid>
<pubDate>Tue, 25 Jul 2023 01:17:09 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=OqfqypXIIwixdRENqBn2sg&_nc_ss=73289&oh=00_Af2_8kSB_zxFVSmfG69Wdg9FkXGf_V-jiMBeLNEx3e_6gQ&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>PyTorch</title>
<description>PyTorch is an open source deep learning framework built to be flexible and modular for research, with the stability and support needed for production deployment.</description>
<link>https://ai.meta.com/tools/pytorch/</link>
<guid isPermaLink="false">297718674239828</guid>
<pubDate>Wed, 08 Mar 2023 02:37:08 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=OqfqypXIIwixdRENqBn2sg&_nc_ss=73289&oh=00_Af2_8kSB_zxFVSmfG69Wdg9FkXGf_V-jiMBeLNEx3e_6gQ&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>Casual Conversations V2 Dataset</title>
<description>Casual Conversations V2 dataset is designed to measure the robustness of AI models across a diverse set of age, genders, apparent skin tones and ambient lighting conditions.</description>
<link>https://ai.meta.com/tools/casual-conversations-v2-dataset/</link>
<guid isPermaLink="false">1646023009169056</guid>
<pubDate>Tue, 07 Mar 2023 23:02:56 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/335035828_927765094902963_5055688398222102698_n.jpg?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=7pNQPMjhOWgQ7kNvwFTxzBc&_nc_oc=Ado3wD7Loft2uKOCf7WdJn1jIwggT7EKrX4pOkxuL9Iw__ofoghxWA5l9mfxswYOHVA&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=OqfqypXIIwixdRENqBn2sg&_nc_ss=73289&oh=00_Af0JVKE-YAI73rLRhFh3yVfD3MdYDhAIkyzn07QKxAleTA&oe=6A08B7BC" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
</channel>
</rss>... |
Contributor
http://localhost:1200/meta/ai/global-search/sort_by=MOST_RECENT - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/sort_by=MOST_RECENT" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/ (sort: MOST_RECENT, total hits: 2848). - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:10:10 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>AIRA₂: Overcoming Bottlenecks in AI Research Agents</title>
<description>Existing research has identified three structural performance bottlenecks in AI research agents: (1) synchronous single-GPU execution constrains sample throughput, limiting the benefit of search; (2) a generalization gap where validation-based selection causes overfitting and performance to degrade over extended search horizons; and (3) the limited capability of fixed, single-turn LLM operators imposes a ceiling on search performance. We introduce AIRA₂, which addresses these bottlenecks through three architectural choices: an asynchronous multi-GPU worker pool that increases experiment throughput linearly; a Hidden Consistent Evaluation protocol that delivers a reliable evaluation signal; and ReAct agents that dynamically scope their actions and debug interactively. On MLE-bench-30, AIRA₂ achieves a mean Percentile Rank of 81.5% at 24 hours and 83.1% at 72 hours, outperforming the strongest baseline, which achieves 72.7%. On AIRS-Bench, AIRA₂ exceeds human state-of-the-art on 6 out of 20 diverse research tasks. Ablations confirm that each architectural component is necessary, that performance follows a predictable scaling law that transfers across LLM backbones, and that the "overfitting" reported in prior work was driven by evaluation noise rather than true data memorization.</description>
<link>https://ai.meta.com/research/publications/aira-overcoming-bottlenecks-in-ai-research-agents/</link>
<guid isPermaLink="false">1298625375513718</guid>
<pubDate>Thu, 16 Apr 2026 07:00:00 GMT</pubDate>
<author>Karen Hambardzumyan, Nicolas Baldwin, Edan Toledo, Rishi Hazra, Michael Kuchnik, Bassel Al Omari, Thomas Simon Foster, Anton Protopopov, Jean-Christophe Gagnon-Audet, Ishita Mediratta, Kelvin Niu, Michael Shvartsman, Alisia Lupidi, Alexis Audran-Reiss, Parth Pathak, Tatiana Shavrina, Despoina Magka, Hela Momand, Derek Dunfield, Nicola Cancedda, Pontus Stenetorp, Carole-Jean Wu, Jakob Foerster, Yoram Bachrach, Martin Josifoski</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
</item>
<item>
<title>TransText: Transparency Aware Image-to-Video Typography Animation</title>
<description>We introduce the first method, to the best of our knowledge, for adapting image-to-video models to layer-aware text (glyph) animation, a capability critical for practical dynamic visual design. Existing approaches predominantly handle the transparency-encoding (alpha channel) as an extra latent dimension appended to the RGB space, necessitating the reconstruction of the underlying RGB-centric variational autoencoder (VAE). However, given the scarcity of high-quality transparent glyph data, retraining the VAE is computationally expensive and may erode the robust semantic priors learned from massive RGB corpora, potentially leading to latent pattern mixing. To mitigate these limitations, we propose TransText, a framework based on a novel Alpha-as-RGB paradigm to jointly model appearance and transparency without modifying the pre-trained generative manifold. TransText embeds the alpha channel as an RGB-compatible visual signal through latent spatial concatenation, explicitly ensuring strict cross-modal (RGB-and-Alpha) consistency while preventing feature entanglement. Our experiments demonstrate that TransText significantly outperforms baselines, generating coherent, high-fidelity transparent animations with diverse, fine-grained effects.</description>
<link>https://ai.meta.com/research/publications/transtext-transparency-aware-image-to-video-typography-animation/</link>
<guid isPermaLink="false">2172593836888174</guid>
<pubDate>Tue, 14 Apr 2026 07:00:00 GMT</pubDate>
<author>Fei Zhang, Zijian Zhou, Bohao Tang, Sen He, Hang Li (BizAI), Zhe Wang, Soubhik Sanyal, Pengfei Liu, Viktar Atliha, Tao Xiang, Frost Xu, Semih Gunel</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Computer Vision</category>
<category>ML Applications</category>
</item>
<item>
<title>Think in Strokes, Not Pixels: Process-Driven Image Generation via Interleaved Reasoning</title>
<description>Humans paint images incrementally: they plan a global layout, sketch a coarse draft, inspect, and refine details, and most importantly, each step is grounded in the evolving visual states. However, can unified multimodal models trained on text-image interleaved datasets also imagine the chain of intermediate states? In this paper, we introduce process-driven image generation, a multi-step paradigm that decomposes synthesis into an interleaved reasoning trajectory of thoughts and actions. Rather than generating images in a single step, our approach unfolds across multiple iterations, each consisting of 4 stages: textual planning, visual drafting, textual reflection, and visual refinement. The textual reasoning explicitly conditions how the visual state should evolve, while the generated visual intermediate in turn constrains and grounds the next round of textual reasoning. A core challenge of process-driven generation stems from the ambiguity of intermediate states: how can models evaluate each partially-complete image? We address this through dense, step-wise supervision that maintains two complementary constraints: for the visual intermediate states, we enforce the spatial and semantic consistency; for the textual intermediate states, we preserve the prior visual knowledge while enabling the model to identify and correct prompt-violating elements. This makes the generation process explicit, interpretable, and directly supervisable. To validate proposed method, we conduct experiments under various text-to-image generation benchmarks.</description>
<link>https://ai.meta.com/research/publications/think-in-strokes-not-pixels-process-driven-image-generation-via-interleaved-reasoning/</link>
<guid isPermaLink="false">847030005080644</guid>
<pubDate>Thu, 09 Apr 2026 07:00:00 GMT</pubDate>
<author>Lei Zhang, Junjiao Tian, Zhipeng Fan, Kunpeng Li, Jialiang Wang, Weifeng Chen, Markos Georgopoulos, Felix Xu, Yuxiao Bao, Julian McAuley, Manling Li, Zecheng He</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Computer Vision</category>
</item>
<item>
<title>Scaling How We Build and Test Our Most Advanced AI</title>
<description>As we build more capable, personalized AI, reliability, security, and user protections are more important than ever. </description>
<link>https://ai.meta.com/blog/scaling-how-we-build-test-advanced-ai/</link>
<guid isPermaLink="false">4369469056623958</guid>
<pubDate>Wed, 08 Apr 2026 17:00:17 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/668130848_1453339552354697_3291493920555502062_n.png?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=yvFttOBoRYkQ7kNvwE2zAxR&_nc_oc=Adr2UifKLIVIje1oDq_d_Zzzd2xhHfN26iDMLvatYMqMAKMUrJtd8QiHDdByuQsgxo8&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2oNbN60GqSv5JZPadi27jXCDAW51Mu-FZVhVDFjuAB_g&oe=6A08CDEE" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Introducing Muse Spark: Scaling Towards Personal Superintelligence</title>
<description></description>
<link>https://ai.meta.com/blog/introducing-muse-spark-msl/</link>
<guid isPermaLink="false">963191949487556</guid>
<pubDate>Wed, 08 Apr 2026 16:00:59 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/667507363_1871749950191919_4996140970371485757_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=IbntS1cBAcsQ7kNvwFGpZpf&_nc_oc=AdoWRIZVlnRJCHKphpg-44LJdWpIcGyseuyTtGVMwVPvYn8IC_2M7OqrJOG1JaQJSk0&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af1ulCi4_xD6KRGqmGzyZB1yOiVUA61MdCK7O3vTYG0dFQ&oe=6A08D22C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>How Alta Daily Uses Meta’s Segment Anything to Reimagine the Digital Closet</title>
<description></description>
<link>https://ai.meta.com/blog/alta-daily-fashion-app-segment-anything/</link>
<guid isPermaLink="false">2194788291268065</guid>
<pubDate>Mon, 06 Apr 2026 18:53:27 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/658967078_993415586351079_325919381939729048_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=cBKF2Wk0XYoQ7kNvwHg8n78&_nc_oc=Adpp8hWFe9GVSkrNsSGgcTK4r7FP6jRCnFsR4SVl4s8_NG3Td8pcFrQWplQHD9AJ0_Y&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af0kQyVclPY4aLmTmg2Gxo9GL9En6rd0yAmT1X7I61S9SQ&oe=6A08B252" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning</title>
<description></description>
<link>https://ai.meta.com/blog/segment-anything-model-3/</link>
<guid isPermaLink="false">715047027731045</guid>
<pubDate>Fri, 27 Mar 2026 16:00:37 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/583339165_852765810541557_5592130597432767690_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=WvtkkpGzh44Q7kNvwEubR7-&_nc_oc=AdosYwdove82TFwfnCLk3q5bXi6OYKYKFlbayU_okqqiX8vtIP70WBbgYw_OqvZ58J8&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af3aIesedlJLWj7riL72uSFGb-VOj_TqcM1dVsBY0CyNCw&oe=6A08D061" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>Introducing TRIBE v2: A Predictive Foundation Model Trained to Understand How the Human Brain Processes Complex Stimuli</title>
<description></description>
<link>https://ai.meta.com/blog/tribe-v2-brain-predictive-foundation-model/</link>
<guid isPermaLink="false">25512119895129580</guid>
<pubDate>Thu, 26 Mar 2026 16:00:16 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/659112495_952140057346056_1728296311373871062_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=rBe_VZxcFD4Q7kNvwGOhdP7&_nc_oc=AdoaYP1BwQzfnCGpSQaVG8V8CaMbbznHgw9z7Mt3CIaeiUr0iUJpZ_p5I6em3s7sWpk&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af25RH3oiq66kUoLuDdKnfb3e-M-OB0AyJwTaZv7tS7lQQ&oe=6A08AF6B" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Research</category>
</item>
<item>
<title>A foundation model of vision, audition, and language for in-silico neuroscience</title>
<description>Cognitive neuroscience is fragmented into specialized models, each tailored to specific experimental paradigms, hence preventing a unified model of cognition in the human brain. Here, we introduce TRIBE v2, a tri-modal (video, audio and language) foundation model capable of predicting human brain activity in a variety of naturalistic and experimental conditions. Leveraging a unified dataset of over 1,000 hours of fMRI across 720 subjects, we demonstrate that our model accurately predicts high-resolution brain responses for novel stimuli, tasks and subjects, superseding traditional linear encoding models, delivering several-fold improvements in accuracy. Critically, TRIBE v2 enables in silico experimentation: tested on seminal visual and neuro-linguistic paradigms, it recovers a variety of results established by decades of empirical research. Finally, by extracting interpretable latent features, TRIBE v2 reveals the fine-grained topography of multisensory integration. These results establish artificial intelligence as a unifying framework for exploring the functional organization of the human brain.</description>
<link>https://ai.meta.com/research/publications/a-foundation-model-of-vision-audition-and-language-for-in-silico-neuroscience/</link>
<guid isPermaLink="false">1009839085553251</guid>
<pubDate>Thu, 26 Mar 2026 07:00:00 GMT</pubDate>
<author>Stéphane d'Ascoli, Jérémy Rapin, Yohann Benchetrit, Teon Brooks, Katelyn Begany, Josephine Raugel, Hubert Jacob Banville, Jean Remi King</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
</item>
<item>
<title>HyperAgents</title>
<description>Self-improving AI systems aim to reduce reliance on human engineering by learning to improve their own learning and problem-solving processes. Existing approaches to recursive self-improvement typically rely on fixed, handcrafted meta-level mechanisms, which fundamentally limit how fast such systems can improve. The Darwin Gödel Machine (DGM)(Zhang et al., 2025b) demonstrates that open-ended self-improvement is achievable in coding. Starting from a single coding agent, the DGM repeatedly generates and evaluates self-modified variants, forming a growing archive of stepping stones for future improvement. Because both evaluation and self-modification are coding tasks, gains in coding ability can translate into gains in self-improvement ability. However, this alignment does not generally hold beyond coding domains.
We introduce hyperagents, self-referential agents that integrate a task agent (which solves the target task) and a meta agent (which modifies itself and the task agent) into a single editable program. Crucially, the meta-level modification procedure is itself editable, enabling metacognitive self-modification, improving not only task-solving behavior, but also the mechanism that generates future improvements. We instantiate this framework by extending DGM to create DGM-Hyperagents (DGM-H). By allowing the improvement procedure to evolve, the DGM-H eliminates the assumption of domain-specific alignment between task performance and self-modification skill, and can potentially support self-accelerating progress on any computable task. Across diverse domains (coding, paper review, robotics reward design, and Olympiad-level math-solution grading), the DGM-H improves performance over time and outperforms baselines without self-improvement or open-ended exploration, as well as prior self-improving systems like DGM. We further show that the DGM-H improves the process by which it generates new agents (e.g., persistent memory, performance tracking), and that these meta-level improvements transfer across domains and accumulate across runs. All experiments were conducted with safety precautions (e.g., sandboxing, human oversight). We discuss what safety entails in this setting and the broader implications of self-improving systems. DGM-Hyperagents offer a glimpse of open-ended AI systems that do not merely search for better solutions, but continually improve their search for how to improve.</description>
<link>https://ai.meta.com/research/publications/hyperagents/</link>
<guid isPermaLink="false">1669124537439877</guid>
<pubDate>Tue, 24 Mar 2026 07:00:00 GMT</pubDate>
<author>Jenny Zhang, Bingchen Zhao, Winnie Yang, Jakob Foerster, Sam Devlin, Tatiana Shavrina</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
<category>Open Source</category>
</item>
<item>
<title>Omnilingual MT: Machine Translation for 1,600 Languages</title>
<description>Advances made through No Language Left Behind (NLLB) have demonstrated that high-quality machine translation (MT) scale to 200 languages. Later Large Language Models (LLMs) have been adopted for MT, increasing in quality but not necessarily extending language coverage. Current systems remain constrained by limited coverage and a persistent generation bottleneck: while crosslingual transfer enables models to somehow understand many undersupported languages, they often cannot generate them reliably, leaving most of the world’s 7,000 languages—especially endangered and marginalized ones—outside the reach of modern MT. Early explorations in extreme scaling offered promising proofs of concept but did not yield sustained solutions.
We present Omnilingual Machine Translation (OMT), the first MT system supporting more than 1,600 languages. This scale is enabled by a comprehensive data strategy that integrates large public multilingual corpora with newly created datasets, including manually curated MeDLEY bitext, synthetic backtranslation, and mining, substantially expanding coverage across long-tail languages, domains, and registers. To ensure both reliable and expansive evaluation, we combined standard metrics with a suite of evaluation artifacts: BLASER 3 quality estimation model (reference-free), OmniTOX toxicity classifier, BOUQuET dataset (a newly created, largest-to-date multilingual evaluation collection built from scratch and manually extended across a wide range of linguistic families), and Met-BOUQuET dataset (faithful multilingual quality estimation at scale). We explore two ways of specializing an LLM for machine translation: as a decoder-only model (OMT-LLaMA) or as a module in an encoder–decoder architecture (OMT-NLLB). The former consists of a model built on LLaMA3, with multilingual continual pretraining and retrieval-augmented translation for inference-time adaptation. The latter is a model built on top of a multilingual aligned space (OmniSONAR, itself also based on LLaMA3), and introduces a training methodology that can exploit non-parallel data, allowing us to incorporate the decoder-only continuous pretraining data into the training of an encoder–decoder architecture. Notably, all our 1B to 8B parameter models match or exceed the MT performance of a 70B LLM baseline, revealing a clear specialization advantage and enabling strong translation quality in low-compute settings. Moreover, our evaluation of English-to-1,600 translations further shows that while baseline models can interpret undersupported languages, they frequently fail to generate them with meaningful fidelity; OMT-LLaMA models substantially expand the set of languages for which coherent generation is feasible. Additionally, OMT models improve in cross-lingual transfer, being close to solving the “understanding” part of the puzzle in MT for the 1,600 evaluated. Beyond strong out-of-the-box performance, we find that finetuning and retrieval-augmented generation offer additional pathways to improve quality for the given subset of languages when targeted data or domain knowledge is available. Our leaderboard and main humanly created evaluation datasets (BOUQuET and Met-BOUQuET) are dynamically evolving towards Omnilinguality and freely available.</description>
<link>https://ai.meta.com/research/publications/omnilingual-mt-machine-translation-for-1600-languages/</link>
<guid isPermaLink="false">1449924233245450</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual MT Team, Belen Alastruey, Niyati Bafna, Andrea Caciolai, Kevin Heffernan, Artyom Kozhevnikov, Christophe Ropers, Eduardo Sánchez, Charles-Eric Saint-James, Ioannis Tsiamas, Chierh CHENG, Joe Chuang, Paul-Ambroise Duquenne, Mark Duppenthaler, Nate Ekberg, Cynthia Gao, Pere Lluís Huguet Cabot, João Maria Janeiro, Jean Maillard, Gabriel Mejia Gonzalez, Holger Schwenk, Edan Toledo, Arina Turkatenko, Albert Ventayol-Boada, Rashel Moritz, Alexandre Mourachko, Surya Parimi, Mary Williamson, Shireen Yates, David Dale, Marta R. Costa-jussa</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Omnilingual SONAR: Cross-Lingual and Cross-Modal Sentence Embeddings Bridging Massively Multilingual Text and Speech</title>
<description>Cross-lingual sentence encoders have traditionally been limited to a few hundred languages, and have sacrificed downstream performance to achieve better alignment across languages, limiting their adoption. In this work, we introduce OmniSONAR, a novel family of omnilingual, cross-lingual and cross-modal sentence embedding models that breaks this barrier. We establish a unified semantic space, natively encompassing text, speech, code and mathematical expressions, while achieving state-of-the-art downstream performance for an unprecedented scale of thousands of languages, from high-resource languages to extremely low-resource varieties.
To achieve this scale without representation collapse and while maintaining top-tier performance in the high-resource languages, we employ a progressive training strategy. We first build a state-of-the-art foundational embedding space for 200 languages using an LLM-initialized Encoder-Decoder, combining token-level decoding with a novel split-softmax contrastive loss and synthetic hard negatives. Leveraging this strong foundational space, we expand to several thousands of language varieties via a specialized two-stage teacher-student encoder distillation framework. Further modeling extensions derived from OmniSONAR address long context inputs and token-centric representations. Finally, we demonstrate the cross-modal extensibility of this space by seamlessly mapping 177 spoken languages into it.
OmniSONAR redefines the state of the art for multilingual representation learning. It halves the cross-lingual similarity search error rate of the previous best models on the 200 languages of FLORES, while also achieving a staggering 15-fold error rate reduction across 1,560 languages in the BIBLE benchmark. Furthermore, our embedding model enables unprecedented translation capabilities, outperforming NLLB-3B on several multilingual benchmarks, and surpassing all previous models, including multi-billion-parameter LLMs, by 15 chrF++ points in 1,560→English translation in the BIBLE benchmark. Beyond alignment and translation, OmniSONAR demonstrates strong general-purpose capabilities across downstream embedding tasks on MTEB and programming languages on XLCoST. For the speech modality, our massively multilingual extension exhibits a 43% lower error rate in cross-lingual and cross-modal similarity search, while achieving 97% of SeamlessM4T performance in speech-to-text translation, despite being a zero-shot translation model trained only with ASR data. Finally, by training an encoder-decoder language model, Spectrum, exclusively on English text that processes OmniSONAR sequences, we unlock immediate high-performance transfer to thousands of languages and the speech modality for complex downstream tasks. These outstanding results position OmniSONAR as a robust, language- and modality-agnostic foundation for any downstream usage.</description>
<link>https://ai.meta.com/research/publications/omnilingual-sonar-cross-lingual-and-cross-modal-sentence-embeddings-bridging-massively-multilingual-text-and-speech/</link>
<guid isPermaLink="false">1231952259116893</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual SONAR Team, João Maria Janeiro, Pere Lluís Huguet Cabot, Ioannis Tsiamas, Yen Meng, Vivek Iyer, Guillem Ramirez, Loic Barrault, Belen Alastruey, Yu-An Chung, Marta R. Costa-jussa, David Dale, Kevin Heffernan, Jaehyeong Jo, Artyom Kozhevnikov, Alexandre Mourachko, Christophe Ropers, Holger Schwenk, Paul-Ambroise Duquenne</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Speech & Audio</category>
</item>
<item>
<title>Four MTIA Chips in Two Years: Scaling AI Experiences for Billions</title>
<description>Serving a wide range of AI models on a global scale, while maintaining the lowest possible costs, is one of the most demanding infrastructure challenges in the industry. </description>
<link>https://ai.meta.com/blog/meta-mtia-scale-ai-chips-for-billions/</link>
<guid isPermaLink="false">814227801000565</guid>
<pubDate>Wed, 11 Mar 2026 14:00:55 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/649719916_2147323009439466_6866394022243367245_n.png?_nc_cat=103&ccb=1-7&_nc_sid=e280be&_nc_ohc=HQUtIT-12s8Q7kNvwHypiBd&_nc_oc=AdqhDC5yIofMP9r2YvlWFr93_aG4GuJ3-iTW5khRJZ89bZxZNaauqtdJK00B6-15sWs&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af3JOC9Z8pbizWm8-nHPQp0PxfGrMae-USdTTp2pe12Cig&oe=6A08CB6C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Mapping the World's Forests with Greater Precision: Introducing Canopy Height Maps v2</title>
<description>In partnership with the World Resources Institute, today we’re announcing Canopy Height Maps v2 (CHMv2), an open source model, along with world-scale maps generated with the model.</description>
<link>https://ai.meta.com/blog/world-resources-institute-dino-canopy-height-maps-v2/</link>
<guid isPermaLink="false">1675459620286410</guid>
<pubDate>Tue, 10 Mar 2026 16:00:06 GMT</pubDate>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.2365-6/644822799_1500169394861918_3213642176014981812_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=5tvZOejkl8MQ7kNvwGvu4MO&_nc_oc=AdrVs4z8wgcKxzuHwX-nVRZ2jwpvpcXnCoqTV1dyBz47sxm5tkNoYpwnT1Jj3LcN8fY&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af134Zy27okj5OkVJpk2whhCibDiKn4yaIqt51DAegpkzA&oe=6A08D606" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
<item>
<title>Unified Vision–Language Modeling via Concept Space Alignment</title>
<description>We introduce v-Sonar, a vision–language embedding space extended from the text-only embedding space Sonar (Omnilingual Embeddings Team et al., 2026), which supports 1500 text languages and 177 speech languages. To construct v-Sonar, we propose a post-hoc alignment pipeline that maps the representations of an existing vision encoder into the Sonar space. We thoroughly evaluate v-Sonar and show that its embeddings achieve competitive performance on text-to-video retrieval. Equipped with the Sonar text decoder, v-Sonar further surpasses state-of-the-art vision–language models on video captioning tasks, including Dream-1k (Bleu 24.3 vs. 19.6) and Vatex (Bleu 45.0 vs. 41.5).
Leveraging v-Sonar, we first demonstrate that the Large Concept Model (LCM; LCM team et al. 2024) operating in Sonar and trained with English text only, can perform both single- and multi-visual concept understanding in a zero-shot manner. Finally, we introduce v-LCM, which extends the LCM with vision–language instruction tuning. v-LCM encodes vision and language inputs into an unified sequence of latent embeddings via v-Sonar and Sonar, and it is trained with the same latent diffusion objective for next-embedding prediction as in LCM’s text-only pre-training. Experiments on a large-scale multilingual and -modal instruction–tuning data mixture highlight the potential of v-LCM: v-LCM matches state-of-the-art vision-language models on tasks covering image/video captioning and question answering, while significantly outperforming them across 61 rich- to low-resource languages out of all 62 tested languages.</description>
<link>https://ai.meta.com/research/publications/unified-vision-language-modeling-via-concept-space-alignment/</link>
<guid isPermaLink="false">1257213679668978</guid>
<pubDate>Fri, 27 Feb 2026 08:00:00 GMT</pubDate>
<author>Yifu Qiu, Paul-Ambroise Duquenne, Holger Schwenk</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Research</category>
</item>
<item>
<title>Learning Personalized Agents from Human Feedback</title>
<description>Modern AI agents are powerful but often fail to align with the idiosyncratic, evolving preferences of individual users. Prior approaches typically rely on static datasets, either training implicit preference models on interaction history or encoding user profiles in external memory. However, these approaches struggle with new users and with preferences that change over time. We introduce Personalized Agents from Human Feedback (PAHF), a framework for continual personalization in which agents learn online from live interaction using explicit per-user memory. PAHF operationalizes a three-step loop: (1) seeking pre-action clarification to resolve ambiguity, (2) grounding actions in preferences retrieved from memory, and (3) integrating post-action feedback to update memory when preferences drift. To evaluate this capability, we develop a four-phase protocol and two benchmarks in embodied manipulation and online shopping. These benchmarks quantify an agent's ability to learn initial preferences from scratch and subsequently adapt to persona shifts. Our theoretical analysis and empirical results show that integrating explicit memory with dual feedback channels is critical: PAHF learns substantially faster and consistently outperforms both no-memory and single-channel baselines, reducing initial personalization error and enabling rapid adaptation to preference shifts.</description>
<link>https://ai.meta.com/research/publications/learning-personalized-agents-from-human-feedback/</link>
<guid isPermaLink="false">910245944695813</guid>
<pubDate>Thu, 26 Feb 2026 08:00:00 GMT</pubDate>
<author>Kaiqu Liang, Julia Kruk, Shengyi Qian, Xianjun Yang, Shengjie Bi, Shaoliang Nie, Michael Zhang, Lijuan Liu, Jaime Fernández Fisac, Shuyan Zhou, Saghar Hosseini</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Conversational AI</category>
<category>Research</category>
</item>
<item>
<title>FERRET: Framework for Expansion Reliant Red Teaming</title>
<description>We introduce a multi-faceted automated red teaming framework in which the goal is to generate multi-modal adversarial conversations that would break a target model and introduce various expansions that would result in more effective and efficient adversarial conversations. The introduced expansions include: 1. Horizontal expansion in which the goal is for the red team model to self-improve and generate more effective conversation starters that would shape a conversation. 2. Vertical expansion in which the goal is to take these conversation starters that are discovered in the horizontal expansion phase and expand them into effective multi-modal conversations and 3. Meta expansion in which the goal is for the red team model to discover more effective multi-modal attack strategies during the course of a conversation. We call our framework FERRET (Framework for Expansion Reliant Red Teaming) and compare it with various existing automated red teaming approaches. In our experiments, we demonstrate the effectiveness of FERRET in generating effective multi-modal adversarial conversations and its superior performance against existing state of the art approaches.</description>
<link>https://ai.meta.com/research/publications/ferret-framework-for-expansion-reliant-red-teaming/</link>
<guid isPermaLink="false">2396003004185464</guid>
<pubDate>Fri, 13 Feb 2026 08:00:00 GMT</pubDate>
<author>Ninareh Mehrabi, Vítor Albiero, Maya Pavlova, Joanna Bitton</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Responsible AI</category>
</item>
<item>
<title>UniT: Unified Multimodal Chain-of-Thought Test-time Scaling</title>
<description>Unified models can handle both multimodal understanding and generation within a single architecture, yet they typically operate in a single pass without iteratively refining their outputs. Many multimodal tasks, especially those involving complex spatial compositions, multiple interacting objects, or evolving instructions, require decomposing instructions, verifying intermediate results, and making iterative corrections. While test-time scaling (TTS) has demonstrated that allocating additional inference compute for iterative reasoning substantially improves language model performance, extending this paradigm to unified multimodal models remains an open challenge.
We introduce UniT, a framework for multimodal chain-of-thought test-time scaling that enables a single unified model to reason, verify, and refine across multiple rounds. UniT combines agentic data synthesis, unified model training, and flexible test-time inference to elicit cognitive behaviors including verification, subgoal decomposition, and content memory. Our key findings are: (1) unified models trained on short reasoning trajectories generalize to longer inference chains at test time; (2) sequential chain-of-thought reasoning provides a more scalable and compute-efficient TTS strategy than parallel sampling; (3) training on generation and editing trajectories improves out-of-distribution visual reasoning. These results establish multimodal test-time scaling as an effective paradigm for advancing both generation and understanding in unified models.</description>
<link>https://ai.meta.com/research/publications/unit-unified-multimodal-chain-of-thought-test-time-scaling/</link>
<guid isPermaLink="false">1477019817147720</guid>
<pubDate>Wed, 11 Feb 2026 08:00:00 GMT</pubDate>
<author>Leon Liangyu Chen, Haoyu Ma, Zhipeng Fan, Ziqi Huang, Animesh Sinha, Xiaoliang Dai, Jialiang Wang, Zecheng He, Jianwei Yang, Chunyuan Li, Junzhe Sun, Chu Wang, Serena Yeung-Levy, Felix Juefei-Xu</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Computer Vision</category>
</item>
<item>
<title>AIRS-Bench: a Suite of Tasks for Frontier AI Research Science Agents</title>
<description>LLM agents hold significant promise for advancing scientific research. To accelerate this progress, we introduce AIRS-Bench (the AI Research Science Benchmark), a suite of 20 tasks sourced from state-of-the-art machine learning papers. These tasks span diverse domains, including language modeling, mathematics, bioinformatics, and time series forecasting. AIRS-Bench tasks assess agentic capabilities over the full research lifecycle---including idea generation, experiment analysis and iterative refinement---without providing baseline code. The AIRS-Bench task format is versatile, enabling easy integration of new tasks and rigorous comparison across different agentic frameworks. We establish baselines using frontier models paired with both sequential and parallel scaffolds. Our results show that agents exceed human SOTA in four tasks but fail to match it in sixteen others. Even when agents surpass human benchmarks, they do not reach the theoretical performance ceiling for the underlying tasks. These findings indicate that AIRS-Bench is far from saturated and offers substantial room for improvement.
We open-source the AIRS-Bench task definitions and evaluation code to catalyze further development in autonomous scientific research.</description>
<link>https://ai.meta.com/research/publications/airs-bench-a-suite-of-tasks-for-frontier-ai-research-science-agents/</link>
<guid isPermaLink="false">2685677328474642</guid>
<pubDate>Tue, 10 Feb 2026 08:00:00 GMT</pubDate>
<author>Alisia Lupidi, Bhavul Gauri, Thomas Simon Foster, Bassel Al Omari, Despoina Magka, Alberto Pepe, Alexis Audran-Reiss, Muna Aghamelu, Nicolas Baldwin, Lucia Cipolina-Kun, Jean-Christophe Gagnon-Audet, Chee Hau Leow, Sandra Lefdal, Hossam Mossalam, Abhinav Moudgil, Saba Nazir, Emanuel Tewolde, Isabel Urrego, Jordi Armengol-Estape, Amar Budhiraja, Gaurav Chaurasia, Abhishek Charnalia, Derek Dunfield, Karen Hambardzumyan, Daniel Izcovich, Martin Josifoski, Ishita Mediratta, Kelvin Niu, Parth Pathak, Michael Shvartsman, Edan Toledo, Anton Protopopov, Roberta Raileanu, Alexander Miller, Tatiana Shavrina, Jakob Foerster, Yoram Bachrach</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2CVzqNnJNQL9ZFIWzirePa54mk5x_P1JHRySxzLQIRMA&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
</item>
<item>
<title>Reducing Government Costs and Increasing Access to Greenspaces in the United Kingdom with DINO</title>
<description>Meta's DINOv2 model is enhancing reforestation efforts around the world. Learn how the UK government is using DINO to help reduce costs and increase access to greenspaces.</description>
<link>https://ai.meta.com/blog/forest-research-dino/</link>
<guid isPermaLink="false">1369350751606207</guid>
<pubDate>Mon, 09 Feb 2026 17:25:45 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/631673937_4911145895833160_8225979506389394021_n.gif?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=N94hGWEZkvUQ7kNvwGqt8rU&_nc_oc=AdokV6EWG3vMd59CgFsHxXNMWZyGwZg3dVfws-Jhaxx2eEfaisVJ7EZypbfXY18vfqk&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af301IuFENvxMEXUwo822B8M-6306GBfkWCerL8H7Z0q3w&oe=6A08CAFF" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
<category>Open Source</category>
</item>
<item>
<title>Multi-room Apartments Simulation (MRAS) Dataset | Meta AI Research</title>
<description>The Multi-Room Apartments Simulation (MRAS) dataset is a multi-modal dataset created for the task of estimating spatially-distributed acoustic parameters in complex scenes.</description>
<link>https://ai.meta.com/datasets/mras-simulated-rooms-dataset/</link>
<guid isPermaLink="false">2247220155683256</guid>
<pubDate>Wed, 28 Jan 2026 18:49:18 GMT</pubDate>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.2365-6/612410014_2220471548441535_606775622330495537_n.jpg?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=1bE1l45d6ZQQ7kNvwGMtrFY&_nc_oc=Adqxxb2bGLw1oq3NG8aZYuMODgbpt__dmsQw5pkrsOQbRgeHEIStu0h09rsY2lXjCgg&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af0gBqOhkH4a4E4FxW6JpaiThi_huhdzzctO7anwQex7tA&oe=6A08CB19" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
<item>
<title>How DINO and SAM are Helping Modernize Essential Medical Triage Practices</title>
<description>By leveraging advanced AI models, teams at the University of Pennsylvania are aiming to bring cutting-edge automation to emergency response.</description>
<link>https://ai.meta.com/blog/upenn-dino-sam-helping-medical-triage/</link>
<guid isPermaLink="false">3255814814579669</guid>
<pubDate>Thu, 18 Dec 2025 18:00:20 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/597680380_857278207264023_6561532616114241333_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=2MKpeQyaIXMQ7kNvwEW_JOV&_nc_oc=Adoz5nmhj_wrxp1GCBlKBrGXATQAKT54gCSnbUJwgRtuQvy7TBeFM8JWoOf4NqzAroI&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af1rERg91ehl6fCGveBGQ5nVCctX8wIJWHW95K0gyHyeew&oe=6A08D437" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>The Universities Space Research Association Applies Segment Anything Model for Responding to Flood Emergencies</title>
<description>The Universities Space Research Association and Meta are collaborating to help support water observing systems set up by the U.S. Geological Survey. </description>
<link>https://ai.meta.com/blog/usra-sam-flood-emergencies/</link>
<guid isPermaLink="false">1381613153671193</guid>
<pubDate>Thu, 18 Dec 2025 18:00:12 GMT</pubDate>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.2365-6/600304853_1519099669397813_6606203512792368025_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=pBY1dQ-wuD4Q7kNvwEtk3Er&_nc_oc=AdqCTI1KUCH0R0iK0pnBK-3F96nsHIHi7HyQscda0R_SXolcNLBmVeOzZ5XnS2bqPMw&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af21sOc8E2XQ6UoP70DjxIUrKM6iKPHU0E3_i4JLh_4d_w&oe=6A08B857" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>How Orakl Oncology is using DINOv2 to accelerate cancer treatment discovery</title>
<description>Orakl Oncology aims to accelerate cancer research and drug development by combining experimental, lab-based insights and machine learning.</description>
<link>https://ai.meta.com/blog/orakl-oncology-dinov2-accelerating-cancer-treatment/</link>
<guid isPermaLink="false">1035491295071881</guid>
<pubDate>Thu, 20 Feb 2025 17:00:31 GMT</pubDate>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/480460718_649637170770478_3994719197111548806_n.jpg?_nc_cat=106&ccb=1-7&_nc_sid=e280be&_nc_ohc=FE_ybvE6IoMQ7kNvwHxpHqx&_nc_oc=AdqKWcB-oYsR3a2eNDId3y_5qtXGIqXAUknFknymdM_3m5mA8aYdrL9y6E2-1jNZVyw&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2E8qI5v2kk_MO17AHhBW925F0UP78gIm9J3yD7vuOzWw&oe=6A08C327" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
<item>
<title>Aaron Defazio</title>
<description>Aaron's research focuses on improving the practice of machine learning through the development of more reliable and theoretically sound methods such as ...</description>
<link>https://ai.meta.com/people/1115638629589333/aaron-defazio/</link>
<guid isPermaLink="false">1115638629589333</guid>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.35477-6/431543048_2088159024878120_1553708304445833368_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=109&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=vd7kajaxkn4Q7kNvwGQ7l84&_nc_oc=AdrVzy9eg4TqCHzxcCZ1L4XI2I740RBa3s9PgGhl0KVDLXIe-YZTwbQUpKKl868Ecec&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af0I7hbG6XN1hIzia1CCD1g07uiqk5L-vYysaCFqHKJX5g&oe=69F449C8" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Speech & Audio</category>
<category>NLP</category>
<category>Theory</category>
<category>Core Machine Learning</category>
</item>
<item>
<title>Abhimanyu Dubey</title>
<description>Abhimanyu is a Research Scientist at Facebook AI Research working on problems in machine learning. He received his Ph.D. in 2021 from the Massachusetts ...</description>
<link>https://ai.meta.com/people/1401136157273652/abhimanyu-dubey/</link>
<guid isPermaLink="false">1401136157273652</guid>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.35477-6/433945609_1344738226210370_2541979406521528116_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=54biatSobnEQ7kNvwHm-noQ&_nc_oc=AdqgN3XWPgzN9zeEi07MRfAbFtwvMz1HwIPWMkl_i0ZncDiOKPKCljvu2nsTH9e6brE&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af08HtP7DtFlGYgdbYuW5tzuY6MJn-NvNbnRVUyihg_9OA&oe=69F4610D" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Computer Vision</category>
<category>Human & Machine Intelligence</category>
<category>Core Machine Learning</category>
<category>Research</category>
<category>ML Applications</category>
</item>
<item>
<title>Abhishek Das</title>
<description>Abhishek is a Research Scientist at Facebook AI Research (FAIR). His research focuses on deep learning and its applications in climate change, and in bu...</description>
<link>https://ai.meta.com/people/401335699167238/abhishek-das/</link>
<guid isPermaLink="false">401335699167238</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431537252_1543342826243118_6010155505132587328_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=102&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=7bKy_GgNvJUQ7kNvwHnc-io&_nc_oc=AdqcGdwhDx2dXwHqsEc4_GeDLVyC-LnWdC1bV1-bWeL-N8jcQe-nKtJ30FPvIdT16SM&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2XnoHnVTG9q9LX7gG0JjnJ_PGDIWI79LMr1djPyQLfPg&oe=69F44FDB" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Computer Vision</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
<category>NLP</category>
</item>
<item>
<title>Adam Polyak</title>
<description>Adam is a research engineer at Facebook AI Research (FAIR), focusing on deep learning, cross domain image generation, audio generation, and voice conver...</description>
<link>https://ai.meta.com/people/1434021433901222/adam-polyak/</link>
<guid isPermaLink="false">1434021433901222</guid>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.35477-6/431518452_1124340545357794_1947310908456905622_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=111&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=qnsH_8koyHsQ7kNvwFb8I9k&_nc_oc=AdqF4anURrXM4aReFHnW4Car9JsZOksY4Hr5eqp9BLYAY_vkyIocVQKUTHBgQv8u84c&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af1d-U19XArsJwfLiF2Zjg36WQ71FKT8G82Vkl1wHLJHXg&oe=69F4377A" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Research</category>
<category>ML Applications</category>
<category>Speech & Audio</category>
</item>
<item>
<title>Adeel Cheema</title>
<description>Adeel is a software engineer working on transparency &amp; control within Meta’s Responsible AI pillar. He focuses on creating product experiences to expl...</description>
<link>https://ai.meta.com/people/1808244339639715/adeel-cheema/</link>
<guid isPermaLink="false">1808244339639715</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431551867_767017845524520_8549663744281985032_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=3dHbSn4toEQQ7kNvwG-hxyL&_nc_oc=AdrneKtcRtDVD2x9eU9c0e9lX-OK623WzkPRI08JRJ3HANhfIiPoSRx5_FjSLcFEWTg&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af3Jo3uOLcAul_KrDJfRvKjRUM-yShqpIxoRWqP2YoAt-Q&oe=69F44D79" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Adina Williams</title>
<description>Adina is a Research Scientist at Facebook AI Research in NYC (started October 2018). Previously, she earned her PhD at New York University in the Depart...</description>
<link>https://ai.meta.com/people/1396973444287406/adina-williams/</link>
<guid isPermaLink="false">1396973444287406</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431521840_2635057386653799_4484141047373535802_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=104&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=D9tP58vAF5kQ7kNvwHvASYA&_nc_oc=AdrKJZSIAfBDaQnc-wW1ToZsOSEhxJpnizTtfwAZJHHOlkTEt1y7LNyQXm2jzG7quUg&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af31WW9SGt1c95w9fhong2H4gNSc3973zdJf1fBgjbLPow&oe=69F467EC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Computer Vision</category>
<category>Core Machine Learning</category>
<category>Open Source</category>
<category>NLP</category>
<category>Conversational AI</category>
<category>Human & Machine Intelligence</category>
<category>ML Applications</category>
</item>
<item>
<title>Adriana Romero Soriano</title>
<description>Adriana is currently a research scientist at Facebook AI Research and an adjunct professor at McGill University. The goal of her research is to develop ...</description>
<link>https://ai.meta.com/people/425710613281765/adriana-romero-soriano/</link>
<guid isPermaLink="false">425710613281765</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431577893_1361257101250742_6288541672250118710_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=nsGSoSWY8rkQ7kNvwFMCuJe&_nc_oc=AdpfuVWkMsXwut0OSpmRwcFwPnozYpICadHJKwLA54PhHlVuRaKtTNxBtL3TcbLP6hk&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af3iEaqGTlJkt902NSIILzgdJJwxKEWBNu9_ESIQnuVX-Q&oe=69F44DCC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Core Machine Learning</category>
<category>Conversational AI</category>
<category>Reinforcement Learni9ng</category>
</item>
<item>
<title>Akhil Mathur</title>
<description>Akhil Mathur is a Research Scientist in the Llama Research team at Meta, working on enhancing the reasoning abilities of Llama models. Previously, as a ...</description>
<link>https://ai.meta.com/people/1251205439201018/akhil-mathur/</link>
<guid isPermaLink="false">1251205439201018</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/457008091_844600247808930_1170475198447093629_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=A2xSui7yUHAQ7kNvwHWOVVV&_nc_oc=Adp4eRq3NJ0287FNM_lETOYqZPQRg2pdbU9cuGT_PdgE67VDPw_dwbYD26U2CO3__R8&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2yaP5A1ez_uBETjtcwtIpXW-fpHz1ml2S3TjXtAib8PA&oe=69F4499C" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Akshara Rai</title>
<description>Akshara is a Research Scientist at Facebook AI Research, working at the intersection of machine learning and control. Her research aims at teaching robo...</description>
<link>https://ai.meta.com/people/423502423531745/akshara-rai/</link>
<guid isPermaLink="false">423502423531745</guid>
<enclosure url="https://scontent-ord5-1.xx.fbcdn.net/v/t39.35477-6/433923962_1381208202581063_6502596142308749692_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=0Wm1dG_MC68Q7kNvwGTCKdY&_nc_oc=AdoD6CrJ1m76HwlDd53FFb_r2h87MuGA8lBXoe2rT1bAb0Kfh-5LqBXoWovSqruL6lU&_nc_zt=14&_nc_ht=scontent-ord5-1.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af0q1MmmGFpeOfwPsjqV1LhPCozln64AoySsnBucpuJPxA&oe=69F43F4F" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Robotics</category>
<category>Computer Vision</category>
<category>Reinforcement Learni9ng</category>
<category>Research</category>
</item>
<item>
<title>Alborz Geramifard</title>
<description>Alborz Geramifard is a senior research manager at Meta AI supporting the Conversational AI. Prior to joining Meta, he led the conversational AI team at ...</description>
<link>https://ai.meta.com/people/363544420007153/alborz-geramifard/</link>
<guid isPermaLink="false">363544420007153</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/431495520_3457483597876936_681607123433531609_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=QzBw61zVxHEQ7kNvwHFGECJ&_nc_oc=Adp_dD05XDi2euaYM6c0WcWx7YkqPn1LQ4QiGQnKLMbD88ygYqi7HyV0szo-h1D2zqM&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af3biQX-9sHWBbyMG9ogKytg-CWbpOn3S-RaIz2wdB11pA&oe=69F45A51" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
<category>Conversational AI</category>
<category>NLP</category>
<category>Computer Vision</category>
</item>
<item>
<title>Alexander H. Miller</title>
<description>Alexander has worked in multi-domain research engineering management, supporting research engineers in London, Tel Aviv, and New York. He is advanced in...</description>
<link>https://ai.meta.com/people/264624513377272/alexander-h-miller/</link>
<guid isPermaLink="false">264624513377272</guid>
<enclosure url="https://scontent-ord5-2.xx.fbcdn.net/v/t39.35477-6/433950826_944833516851120_841662693953017766_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=6ftFVCshb-8Q7kNvwFmW3Vx&_nc_oc=AdpYZTkVT_RFDrpt-_sscuIcqOYLysabf-d4BJfV3EnXsHNOYkP_SRmSHFKxebGWjaU&_nc_zt=14&_nc_ht=scontent-ord5-2.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2Luh9SwNGSbaF16QyM5IdxnEzbxwbE6x1aHkxxja0wpA&oe=69F45C6C" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Reinforcement Learni9ng</category>
<category>Graphics</category>
<category>NLP</category>
</item>
<item>
<title>Amey Porobo Dharwadker</title>
<description>Amey Porobo Dharwadker works as a Machine Learning Engineering Manager, leading Facebook's Video Recommendations Ranking team. Renowned for his expertis...</description>
<link>https://ai.meta.com/people/393924029921120/amey-porobo-dharwadker/</link>
<guid isPermaLink="false">393924029921120</guid>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.35477-6/431510720_7490594577724788_4289788014840147933_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=106&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=CDFvi0iDZRcQ7kNvwFhizrj&_nc_oc=Adr9sJsLUd39ruFsTcvrZXX1i_tYXLf-EyYjs5A_aQ4ACUVT9oNSSPcZpMVShrui3pk&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=dSg1wqAf6fFjTIF_Svirlg&_nc_ss=73289&oh=00_Af2OzoW4Rjz_OrFkFBDgVBG-nShqvh7qCdY8337mB3em8A&oe=69F440F5" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
</channel>
</rss>http://localhost:1200/meta/ai/global-search/q=llama%26content_types=publication%26years=2024,2025%26sort_by=MOST_RECENT - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — q=llama · content_types=publication · years=2024,2025</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/q=llama%26content_types=publication%26years=2024,2025%26sort_by=MOST_RECENT" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/ (sort: MOST_RECENT, total hits: 2). - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:10:14 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>Accelerating a Triton Fused Kernel for W4A16 Quantized Inference with SplitK Work Decomposition</title>
<description>We propose an implementation of an efficient fused matrix multiplication kernel for W4A16 quantized inference, where we perform dequantization and GEMM in a fused kernel using a SplitK work decomposition.
Our implementation shows improvement for the type of skinny matrix-matrix multiplications found in foundation model inference workloads. In particular, this paper surveys the type of matrix multiplication between a skinny activation matrix and a square weight matrix. Our results show an average of 65\% speed improvement on A100, and an average of 124\% speed improvement on H100 (with a peak of 295\%) for a range of matrix dimensions including those found in a llama-style model, where m &lt; n = k.</description>
<link>https://ai.meta.com/research/publications/accelerating-a-triton-fused-kernel-for-w4a16-quantized-inference-with-splitk-work-decomposition/</link>
<guid isPermaLink="false">347040654790634</guid>
<pubDate>Tue, 09 Jan 2024 16:00:00 GMT</pubDate>
<author>Less Wright, Adnan Hoque</author>
<enclosure url="https://scontent-ord5-3.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwEtuFlQ&_nc_oc=Adpqep7pNFwGkWY7kTkbiPbHwN-HK24BbhwiEqvfatIZxWJyhJ1DrJID2L1TTRBqyQQ&_nc_zt=14&_nc_ht=scontent-ord5-3.xx&_nc_gid=-gOM5CuLgy9JhygeK-6nbA&_nc_ss=73289&oh=00_Af1MxlEq2xliLKnmysqCATSgP_2A14KfieifoUI65QbORw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Core Machine Learning</category>
</item>
</channel>
</rss>... |
- Drop `limit` from `parameters` (Rule 7); only path params belong there. Move the `limit` hint into the route's `description` field. - Strip dynamic search metadata (`sort`, `total hits`) from the feed `description` (Rule 11). Keep it as a static description of the feed.
Contributor
Auto ReviewNo clear rule violations found in the current diff. |
Contributor
|
Successfully generated as following: http://localhost:1200/meta/ai/global-search - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/. - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:22:22 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>AIRA₂: Overcoming Bottlenecks in AI Research Agents</title>
<description>Existing research has identified three structural performance bottlenecks in AI research agents: (1) synchronous single-GPU execution constrains sample throughput, limiting the benefit of search; (2) a generalization gap where validation-based selection causes overfitting and performance to degrade over extended search horizons; and (3) the limited capability of fixed, single-turn LLM operators imposes a ceiling on search performance. We introduce AIRA₂, which addresses these bottlenecks through three architectural choices: an asynchronous multi-GPU worker pool that increases experiment throughput linearly; a Hidden Consistent Evaluation protocol that delivers a reliable evaluation signal; and ReAct agents that dynamically scope their actions and debug interactively. On MLE-bench-30, AIRA₂ achieves a mean Percentile Rank of 81.5% at 24 hours and 83.1% at 72 hours, outperforming the strongest baseline, which achieves 72.7%. On AIRS-Bench, AIRA₂ exceeds human state-of-the-art on 6 out of 20 diverse research tasks. Ablations confirm that each architectural component is necessary, that performance follows a predictable scaling law that transfers across LLM backbones, and that the "overfitting" reported in prior work was driven by evaluation noise rather than true data memorization.</description>
<link>https://ai.meta.com/research/publications/aira-overcoming-bottlenecks-in-ai-research-agents/</link>
<guid isPermaLink="false">1298625375513718</guid>
<pubDate>Thu, 16 Apr 2026 07:00:00 GMT</pubDate>
<author>Karen Hambardzumyan, Nicolas Baldwin, Edan Toledo, Rishi Hazra, Michael Kuchnik, Bassel Al Omari, Thomas Simon Foster, Anton Protopopov, Jean-Christophe Gagnon-Audet, Ishita Mediratta, Kelvin Niu, Michael Shvartsman, Alisia Lupidi, Alexis Audran-Reiss, Parth Pathak, Tatiana Shavrina, Despoina Magka, Hela Momand, Derek Dunfield, Nicola Cancedda, Pontus Stenetorp, Carole-Jean Wu, Jakob Foerster, Yoram Bachrach, Martin Josifoski</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
</item>
<item>
<title>TransText: Transparency Aware Image-to-Video Typography Animation</title>
<description>We introduce the first method, to the best of our knowledge, for adapting image-to-video models to layer-aware text (glyph) animation, a capability critical for practical dynamic visual design. Existing approaches predominantly handle the transparency-encoding (alpha channel) as an extra latent dimension appended to the RGB space, necessitating the reconstruction of the underlying RGB-centric variational autoencoder (VAE). However, given the scarcity of high-quality transparent glyph data, retraining the VAE is computationally expensive and may erode the robust semantic priors learned from massive RGB corpora, potentially leading to latent pattern mixing. To mitigate these limitations, we propose TransText, a framework based on a novel Alpha-as-RGB paradigm to jointly model appearance and transparency without modifying the pre-trained generative manifold. TransText embeds the alpha channel as an RGB-compatible visual signal through latent spatial concatenation, explicitly ensuring strict cross-modal (RGB-and-Alpha) consistency while preventing feature entanglement. Our experiments demonstrate that TransText significantly outperforms baselines, generating coherent, high-fidelity transparent animations with diverse, fine-grained effects.</description>
<link>https://ai.meta.com/research/publications/transtext-transparency-aware-image-to-video-typography-animation/</link>
<guid isPermaLink="false">2172593836888174</guid>
<pubDate>Tue, 14 Apr 2026 07:00:00 GMT</pubDate>
<author>Fei Zhang, Zijian Zhou, Bohao Tang, Sen He, Hang Li (BizAI), Zhe Wang, Soubhik Sanyal, Pengfei Liu, Viktar Atliha, Tao Xiang, Frost Xu, Semih Gunel</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Computer Vision</category>
<category>ML Applications</category>
</item>
<item>
<title>Think in Strokes, Not Pixels: Process-Driven Image Generation via Interleaved Reasoning</title>
<description>Humans paint images incrementally: they plan a global layout, sketch a coarse draft, inspect, and refine details, and most importantly, each step is grounded in the evolving visual states. However, can unified multimodal models trained on text-image interleaved datasets also imagine the chain of intermediate states? In this paper, we introduce process-driven image generation, a multi-step paradigm that decomposes synthesis into an interleaved reasoning trajectory of thoughts and actions. Rather than generating images in a single step, our approach unfolds across multiple iterations, each consisting of 4 stages: textual planning, visual drafting, textual reflection, and visual refinement. The textual reasoning explicitly conditions how the visual state should evolve, while the generated visual intermediate in turn constrains and grounds the next round of textual reasoning. A core challenge of process-driven generation stems from the ambiguity of intermediate states: how can models evaluate each partially-complete image? We address this through dense, step-wise supervision that maintains two complementary constraints: for the visual intermediate states, we enforce the spatial and semantic consistency; for the textual intermediate states, we preserve the prior visual knowledge while enabling the model to identify and correct prompt-violating elements. This makes the generation process explicit, interpretable, and directly supervisable. To validate proposed method, we conduct experiments under various text-to-image generation benchmarks.</description>
<link>https://ai.meta.com/research/publications/think-in-strokes-not-pixels-process-driven-image-generation-via-interleaved-reasoning/</link>
<guid isPermaLink="false">847030005080644</guid>
<pubDate>Thu, 09 Apr 2026 07:00:00 GMT</pubDate>
<author>Lei Zhang, Junjiao Tian, Zhipeng Fan, Kunpeng Li, Jialiang Wang, Weifeng Chen, Markos Georgopoulos, Felix Xu, Yuxiao Bao, Julian McAuley, Manling Li, Zecheng He</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Computer Vision</category>
</item>
<item>
<title>Scaling How We Build and Test Our Most Advanced AI</title>
<description>As we build more capable, personalized AI, reliability, security, and user protections are more important than ever. </description>
<link>https://ai.meta.com/blog/scaling-how-we-build-test-advanced-ai/</link>
<guid isPermaLink="false">4369469056623958</guid>
<pubDate>Wed, 08 Apr 2026 17:00:17 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/668130848_1453339552354697_3291493920555502062_n.png?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=yvFttOBoRYkQ7kNvwHAGoUI&_nc_oc=AdrMTKarV4pYb-C6Lo7cZ4FS8VtfHNic415DSbyvO9pcVHrt8wGk3pW7eGF_3ATLqB8&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af2btXyVpUDWqE83ipdr2-JAdjHr_UOnJ-LBrsuwm-HuyQ&oe=6A08CDEE" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Introducing Muse Spark: Scaling Towards Personal Superintelligence</title>
<description></description>
<link>https://ai.meta.com/blog/introducing-muse-spark-msl/</link>
<guid isPermaLink="false">963191949487556</guid>
<pubDate>Wed, 08 Apr 2026 16:00:59 GMT</pubDate>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.2365-6/667507363_1871749950191919_4996140970371485757_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=IbntS1cBAcsQ7kNvwEkZk-q&_nc_oc=Ado4EZgw05h6efdlgHKUmEWqiQAUivkkK5wab1s8qYcxECiinRzqsBcSp7BtuIVOo5k&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af3y1m-WFgM249z2vZx3ep245KIq_5h96T7KDATt2flypA&oe=6A08D22C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>How Alta Daily Uses Meta’s Segment Anything to Reimagine the Digital Closet</title>
<description></description>
<link>https://ai.meta.com/blog/alta-daily-fashion-app-segment-anything/</link>
<guid isPermaLink="false">2194788291268065</guid>
<pubDate>Mon, 06 Apr 2026 18:53:27 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/658967078_993415586351079_325919381939729048_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=cBKF2Wk0XYoQ7kNvwGT68WW&_nc_oc=Adr_OUg82eUw0EbluNihH00lRvwhw0cI82roesUhcZVs9_YWPOnE24LdOGDODhZ8PXE&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af2-PvInRb-o03kqN8ovcdg-DPrdrVFP9D4M5Y1po4xXWQ&oe=6A08B252" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning</title>
<description></description>
<link>https://ai.meta.com/blog/segment-anything-model-3/</link>
<guid isPermaLink="false">715047027731045</guid>
<pubDate>Fri, 27 Mar 2026 16:00:37 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/583339165_852765810541557_5592130597432767690_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=WvtkkpGzh44Q7kNvwHtq_BU&_nc_oc=AdqOSUbFE1HjLhxvY58CCZUsdmat-y6v9A-cIg_tkySMD-A2NT_eE5pWvyzi6Gtrlq0&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af3JxCJtgrukU-So_hweqITShlaz0b-ATB5MwT0EWbmxow&oe=6A08D061" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>Introducing TRIBE v2: A Predictive Foundation Model Trained to Understand How the Human Brain Processes Complex Stimuli</title>
<description></description>
<link>https://ai.meta.com/blog/tribe-v2-brain-predictive-foundation-model/</link>
<guid isPermaLink="false">25512119895129580</guid>
<pubDate>Thu, 26 Mar 2026 16:00:16 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/659112495_952140057346056_1728296311373871062_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=rBe_VZxcFD4Q7kNvwEl0TGK&_nc_oc=Adq6bzVjbryLAhis8bGJPU1Umwebg8J3Ul_Sn-xoSPOJfZTVx6fNEnrsqhxPP2J05wc&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af3A0dK5PK9hzj9jrZPZCMpE1fprbVC2az4eapCkLPOdzg&oe=6A08AF6B" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Research</category>
</item>
<item>
<title>A foundation model of vision, audition, and language for in-silico neuroscience</title>
<description>Cognitive neuroscience is fragmented into specialized models, each tailored to specific experimental paradigms, hence preventing a unified model of cognition in the human brain. Here, we introduce TRIBE v2, a tri-modal (video, audio and language) foundation model capable of predicting human brain activity in a variety of naturalistic and experimental conditions. Leveraging a unified dataset of over 1,000 hours of fMRI across 720 subjects, we demonstrate that our model accurately predicts high-resolution brain responses for novel stimuli, tasks and subjects, superseding traditional linear encoding models, delivering several-fold improvements in accuracy. Critically, TRIBE v2 enables in silico experimentation: tested on seminal visual and neuro-linguistic paradigms, it recovers a variety of results established by decades of empirical research. Finally, by extracting interpretable latent features, TRIBE v2 reveals the fine-grained topography of multisensory integration. These results establish artificial intelligence as a unifying framework for exploring the functional organization of the human brain.</description>
<link>https://ai.meta.com/research/publications/a-foundation-model-of-vision-audition-and-language-for-in-silico-neuroscience/</link>
<guid isPermaLink="false">1009839085553251</guid>
<pubDate>Thu, 26 Mar 2026 07:00:00 GMT</pubDate>
<author>Stéphane d'Ascoli, Jérémy Rapin, Yohann Benchetrit, Teon Brooks, Katelyn Begany, Josephine Raugel, Hubert Jacob Banville, Jean Remi King</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
</item>
<item>
<title>HyperAgents</title>
<description>Self-improving AI systems aim to reduce reliance on human engineering by learning to improve their own learning and problem-solving processes. Existing approaches to recursive self-improvement typically rely on fixed, handcrafted meta-level mechanisms, which fundamentally limit how fast such systems can improve. The Darwin Gödel Machine (DGM)(Zhang et al., 2025b) demonstrates that open-ended self-improvement is achievable in coding. Starting from a single coding agent, the DGM repeatedly generates and evaluates self-modified variants, forming a growing archive of stepping stones for future improvement. Because both evaluation and self-modification are coding tasks, gains in coding ability can translate into gains in self-improvement ability. However, this alignment does not generally hold beyond coding domains.
We introduce hyperagents, self-referential agents that integrate a task agent (which solves the target task) and a meta agent (which modifies itself and the task agent) into a single editable program. Crucially, the meta-level modification procedure is itself editable, enabling metacognitive self-modification, improving not only task-solving behavior, but also the mechanism that generates future improvements. We instantiate this framework by extending DGM to create DGM-Hyperagents (DGM-H). By allowing the improvement procedure to evolve, the DGM-H eliminates the assumption of domain-specific alignment between task performance and self-modification skill, and can potentially support self-accelerating progress on any computable task. Across diverse domains (coding, paper review, robotics reward design, and Olympiad-level math-solution grading), the DGM-H improves performance over time and outperforms baselines without self-improvement or open-ended exploration, as well as prior self-improving systems like DGM. We further show that the DGM-H improves the process by which it generates new agents (e.g., persistent memory, performance tracking), and that these meta-level improvements transfer across domains and accumulate across runs. All experiments were conducted with safety precautions (e.g., sandboxing, human oversight). We discuss what safety entails in this setting and the broader implications of self-improving systems. DGM-Hyperagents offer a glimpse of open-ended AI systems that do not merely search for better solutions, but continually improve their search for how to improve.</description>
<link>https://ai.meta.com/research/publications/hyperagents/</link>
<guid isPermaLink="false">1669124537439877</guid>
<pubDate>Tue, 24 Mar 2026 07:00:00 GMT</pubDate>
<author>Jenny Zhang, Bingchen Zhao, Winnie Yang, Jakob Foerster, Sam Devlin, Tatiana Shavrina</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
<category>Open Source</category>
</item>
<item>
<title>Omnilingual MT: Machine Translation for 1,600 Languages</title>
<description>Advances made through No Language Left Behind (NLLB) have demonstrated that high-quality machine translation (MT) scale to 200 languages. Later Large Language Models (LLMs) have been adopted for MT, increasing in quality but not necessarily extending language coverage. Current systems remain constrained by limited coverage and a persistent generation bottleneck: while crosslingual transfer enables models to somehow understand many undersupported languages, they often cannot generate them reliably, leaving most of the world’s 7,000 languages—especially endangered and marginalized ones—outside the reach of modern MT. Early explorations in extreme scaling offered promising proofs of concept but did not yield sustained solutions.
We present Omnilingual Machine Translation (OMT), the first MT system supporting more than 1,600 languages. This scale is enabled by a comprehensive data strategy that integrates large public multilingual corpora with newly created datasets, including manually curated MeDLEY bitext, synthetic backtranslation, and mining, substantially expanding coverage across long-tail languages, domains, and registers. To ensure both reliable and expansive evaluation, we combined standard metrics with a suite of evaluation artifacts: BLASER 3 quality estimation model (reference-free), OmniTOX toxicity classifier, BOUQuET dataset (a newly created, largest-to-date multilingual evaluation collection built from scratch and manually extended across a wide range of linguistic families), and Met-BOUQuET dataset (faithful multilingual quality estimation at scale). We explore two ways of specializing an LLM for machine translation: as a decoder-only model (OMT-LLaMA) or as a module in an encoder–decoder architecture (OMT-NLLB). The former consists of a model built on LLaMA3, with multilingual continual pretraining and retrieval-augmented translation for inference-time adaptation. The latter is a model built on top of a multilingual aligned space (OmniSONAR, itself also based on LLaMA3), and introduces a training methodology that can exploit non-parallel data, allowing us to incorporate the decoder-only continuous pretraining data into the training of an encoder–decoder architecture. Notably, all our 1B to 8B parameter models match or exceed the MT performance of a 70B LLM baseline, revealing a clear specialization advantage and enabling strong translation quality in low-compute settings. Moreover, our evaluation of English-to-1,600 translations further shows that while baseline models can interpret undersupported languages, they frequently fail to generate them with meaningful fidelity; OMT-LLaMA models substantially expand the set of languages for which coherent generation is feasible. Additionally, OMT models improve in cross-lingual transfer, being close to solving the “understanding” part of the puzzle in MT for the 1,600 evaluated. Beyond strong out-of-the-box performance, we find that finetuning and retrieval-augmented generation offer additional pathways to improve quality for the given subset of languages when targeted data or domain knowledge is available. Our leaderboard and main humanly created evaluation datasets (BOUQuET and Met-BOUQuET) are dynamically evolving towards Omnilinguality and freely available.</description>
<link>https://ai.meta.com/research/publications/omnilingual-mt-machine-translation-for-1600-languages/</link>
<guid isPermaLink="false">1449924233245450</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual MT Team, Belen Alastruey, Niyati Bafna, Andrea Caciolai, Kevin Heffernan, Artyom Kozhevnikov, Christophe Ropers, Eduardo Sánchez, Charles-Eric Saint-James, Ioannis Tsiamas, Chierh CHENG, Joe Chuang, Paul-Ambroise Duquenne, Mark Duppenthaler, Nate Ekberg, Cynthia Gao, Pere Lluís Huguet Cabot, João Maria Janeiro, Jean Maillard, Gabriel Mejia Gonzalez, Holger Schwenk, Edan Toledo, Arina Turkatenko, Albert Ventayol-Boada, Rashel Moritz, Alexandre Mourachko, Surya Parimi, Mary Williamson, Shireen Yates, David Dale, Marta R. Costa-jussa</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Omnilingual SONAR: Cross-Lingual and Cross-Modal Sentence Embeddings Bridging Massively Multilingual Text and Speech</title>
<description>Cross-lingual sentence encoders have traditionally been limited to a few hundred languages, and have sacrificed downstream performance to achieve better alignment across languages, limiting their adoption. In this work, we introduce OmniSONAR, a novel family of omnilingual, cross-lingual and cross-modal sentence embedding models that breaks this barrier. We establish a unified semantic space, natively encompassing text, speech, code and mathematical expressions, while achieving state-of-the-art downstream performance for an unprecedented scale of thousands of languages, from high-resource languages to extremely low-resource varieties.
To achieve this scale without representation collapse and while maintaining top-tier performance in the high-resource languages, we employ a progressive training strategy. We first build a state-of-the-art foundational embedding space for 200 languages using an LLM-initialized Encoder-Decoder, combining token-level decoding with a novel split-softmax contrastive loss and synthetic hard negatives. Leveraging this strong foundational space, we expand to several thousands of language varieties via a specialized two-stage teacher-student encoder distillation framework. Further modeling extensions derived from OmniSONAR address long context inputs and token-centric representations. Finally, we demonstrate the cross-modal extensibility of this space by seamlessly mapping 177 spoken languages into it.
OmniSONAR redefines the state of the art for multilingual representation learning. It halves the cross-lingual similarity search error rate of the previous best models on the 200 languages of FLORES, while also achieving a staggering 15-fold error rate reduction across 1,560 languages in the BIBLE benchmark. Furthermore, our embedding model enables unprecedented translation capabilities, outperforming NLLB-3B on several multilingual benchmarks, and surpassing all previous models, including multi-billion-parameter LLMs, by 15 chrF++ points in 1,560→English translation in the BIBLE benchmark. Beyond alignment and translation, OmniSONAR demonstrates strong general-purpose capabilities across downstream embedding tasks on MTEB and programming languages on XLCoST. For the speech modality, our massively multilingual extension exhibits a 43% lower error rate in cross-lingual and cross-modal similarity search, while achieving 97% of SeamlessM4T performance in speech-to-text translation, despite being a zero-shot translation model trained only with ASR data. Finally, by training an encoder-decoder language model, Spectrum, exclusively on English text that processes OmniSONAR sequences, we unlock immediate high-performance transfer to thousands of languages and the speech modality for complex downstream tasks. These outstanding results position OmniSONAR as a robust, language- and modality-agnostic foundation for any downstream usage.</description>
<link>https://ai.meta.com/research/publications/omnilingual-sonar-cross-lingual-and-cross-modal-sentence-embeddings-bridging-massively-multilingual-text-and-speech/</link>
<guid isPermaLink="false">1231952259116893</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual SONAR Team, João Maria Janeiro, Pere Lluís Huguet Cabot, Ioannis Tsiamas, Yen Meng, Vivek Iyer, Guillem Ramirez, Loic Barrault, Belen Alastruey, Yu-An Chung, Marta R. Costa-jussa, David Dale, Kevin Heffernan, Jaehyeong Jo, Artyom Kozhevnikov, Alexandre Mourachko, Christophe Ropers, Holger Schwenk, Paul-Ambroise Duquenne</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Speech & Audio</category>
</item>
<item>
<title>Four MTIA Chips in Two Years: Scaling AI Experiences for Billions</title>
<description>Serving a wide range of AI models on a global scale, while maintaining the lowest possible costs, is one of the most demanding infrastructure challenges in the industry. </description>
<link>https://ai.meta.com/blog/meta-mtia-scale-ai-chips-for-billions/</link>
<guid isPermaLink="false">814227801000565</guid>
<pubDate>Wed, 11 Mar 2026 14:00:55 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/649719916_2147323009439466_6866394022243367245_n.png?_nc_cat=103&ccb=1-7&_nc_sid=e280be&_nc_ohc=HQUtIT-12s8Q7kNvwGOLZwg&_nc_oc=AdrxxwjcxH_agF9UMFF-AQWK2vy12eCF5p5jf1ozY-qI61a_wgoP7vvJizTyzRxV0xc&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0LvyEX84eOqrYhhSf7LNBJy4Vw-P23g5X4yvDnlLXn5A&oe=6A08CB6C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Mapping the World's Forests with Greater Precision: Introducing Canopy Height Maps v2</title>
<description>In partnership with the World Resources Institute, today we’re announcing Canopy Height Maps v2 (CHMv2), an open source model, along with world-scale maps generated with the model.</description>
<link>https://ai.meta.com/blog/world-resources-institute-dino-canopy-height-maps-v2/</link>
<guid isPermaLink="false">1675459620286410</guid>
<pubDate>Tue, 10 Mar 2026 16:00:06 GMT</pubDate>
<enclosure url="https://scontent-hkg4-2.xx.fbcdn.net/v/t39.2365-6/644822799_1500169394861918_3213642176014981812_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=5tvZOejkl8MQ7kNvwE3rzOr&_nc_oc=AdpUWj25o490QP_XxwEWiK2CqMWOSxEQWHDJ-K-g7z3HzWbeYKo0AeH72yUz_8SC2Pk&_nc_zt=14&_nc_ht=scontent-hkg4-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af3rfVOt71lXZHQzVf20gRBb7QAuTP9hbN7mLWlejC_rng&oe=6A08D606" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
<item>
<title>Unified Vision–Language Modeling via Concept Space Alignment</title>
<description>We introduce v-Sonar, a vision–language embedding space extended from the text-only embedding space Sonar (Omnilingual Embeddings Team et al., 2026), which supports 1500 text languages and 177 speech languages. To construct v-Sonar, we propose a post-hoc alignment pipeline that maps the representations of an existing vision encoder into the Sonar space. We thoroughly evaluate v-Sonar and show that its embeddings achieve competitive performance on text-to-video retrieval. Equipped with the Sonar text decoder, v-Sonar further surpasses state-of-the-art vision–language models on video captioning tasks, including Dream-1k (Bleu 24.3 vs. 19.6) and Vatex (Bleu 45.0 vs. 41.5).
Leveraging v-Sonar, we first demonstrate that the Large Concept Model (LCM; LCM team et al. 2024) operating in Sonar and trained with English text only, can perform both single- and multi-visual concept understanding in a zero-shot manner. Finally, we introduce v-LCM, which extends the LCM with vision–language instruction tuning. v-LCM encodes vision and language inputs into an unified sequence of latent embeddings via v-Sonar and Sonar, and it is trained with the same latent diffusion objective for next-embedding prediction as in LCM’s text-only pre-training. Experiments on a large-scale multilingual and -modal instruction–tuning data mixture highlight the potential of v-LCM: v-LCM matches state-of-the-art vision-language models on tasks covering image/video captioning and question answering, while significantly outperforming them across 61 rich- to low-resource languages out of all 62 tested languages.</description>
<link>https://ai.meta.com/research/publications/unified-vision-language-modeling-via-concept-space-alignment/</link>
<guid isPermaLink="false">1257213679668978</guid>
<pubDate>Fri, 27 Feb 2026 08:00:00 GMT</pubDate>
<author>Yifu Qiu, Paul-Ambroise Duquenne, Holger Schwenk</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Research</category>
</item>
<item>
<title>Learning Personalized Agents from Human Feedback</title>
<description>Modern AI agents are powerful but often fail to align with the idiosyncratic, evolving preferences of individual users. Prior approaches typically rely on static datasets, either training implicit preference models on interaction history or encoding user profiles in external memory. However, these approaches struggle with new users and with preferences that change over time. We introduce Personalized Agents from Human Feedback (PAHF), a framework for continual personalization in which agents learn online from live interaction using explicit per-user memory. PAHF operationalizes a three-step loop: (1) seeking pre-action clarification to resolve ambiguity, (2) grounding actions in preferences retrieved from memory, and (3) integrating post-action feedback to update memory when preferences drift. To evaluate this capability, we develop a four-phase protocol and two benchmarks in embodied manipulation and online shopping. These benchmarks quantify an agent's ability to learn initial preferences from scratch and subsequently adapt to persona shifts. Our theoretical analysis and empirical results show that integrating explicit memory with dual feedback channels is critical: PAHF learns substantially faster and consistently outperforms both no-memory and single-channel baselines, reducing initial personalization error and enabling rapid adaptation to preference shifts.</description>
<link>https://ai.meta.com/research/publications/learning-personalized-agents-from-human-feedback/</link>
<guid isPermaLink="false">910245944695813</guid>
<pubDate>Thu, 26 Feb 2026 08:00:00 GMT</pubDate>
<author>Kaiqu Liang, Julia Kruk, Shengyi Qian, Xianjun Yang, Shengjie Bi, Shaoliang Nie, Michael Zhang, Lijuan Liu, Jaime Fernández Fisac, Shuyan Zhou, Saghar Hosseini</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Conversational AI</category>
<category>Research</category>
</item>
<item>
<title>FERRET: Framework for Expansion Reliant Red Teaming</title>
<description>We introduce a multi-faceted automated red teaming framework in which the goal is to generate multi-modal adversarial conversations that would break a target model and introduce various expansions that would result in more effective and efficient adversarial conversations. The introduced expansions include: 1. Horizontal expansion in which the goal is for the red team model to self-improve and generate more effective conversation starters that would shape a conversation. 2. Vertical expansion in which the goal is to take these conversation starters that are discovered in the horizontal expansion phase and expand them into effective multi-modal conversations and 3. Meta expansion in which the goal is for the red team model to discover more effective multi-modal attack strategies during the course of a conversation. We call our framework FERRET (Framework for Expansion Reliant Red Teaming) and compare it with various existing automated red teaming approaches. In our experiments, we demonstrate the effectiveness of FERRET in generating effective multi-modal adversarial conversations and its superior performance against existing state of the art approaches.</description>
<link>https://ai.meta.com/research/publications/ferret-framework-for-expansion-reliant-red-teaming/</link>
<guid isPermaLink="false">2396003004185464</guid>
<pubDate>Fri, 13 Feb 2026 08:00:00 GMT</pubDate>
<author>Ninareh Mehrabi, Vítor Albiero, Maya Pavlova, Joanna Bitton</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Responsible AI</category>
</item>
<item>
<title>UniT: Unified Multimodal Chain-of-Thought Test-time Scaling</title>
<description>Unified models can handle both multimodal understanding and generation within a single architecture, yet they typically operate in a single pass without iteratively refining their outputs. Many multimodal tasks, especially those involving complex spatial compositions, multiple interacting objects, or evolving instructions, require decomposing instructions, verifying intermediate results, and making iterative corrections. While test-time scaling (TTS) has demonstrated that allocating additional inference compute for iterative reasoning substantially improves language model performance, extending this paradigm to unified multimodal models remains an open challenge.
We introduce UniT, a framework for multimodal chain-of-thought test-time scaling that enables a single unified model to reason, verify, and refine across multiple rounds. UniT combines agentic data synthesis, unified model training, and flexible test-time inference to elicit cognitive behaviors including verification, subgoal decomposition, and content memory. Our key findings are: (1) unified models trained on short reasoning trajectories generalize to longer inference chains at test time; (2) sequential chain-of-thought reasoning provides a more scalable and compute-efficient TTS strategy than parallel sampling; (3) training on generation and editing trajectories improves out-of-distribution visual reasoning. These results establish multimodal test-time scaling as an effective paradigm for advancing both generation and understanding in unified models.</description>
<link>https://ai.meta.com/research/publications/unit-unified-multimodal-chain-of-thought-test-time-scaling/</link>
<guid isPermaLink="false">1477019817147720</guid>
<pubDate>Wed, 11 Feb 2026 08:00:00 GMT</pubDate>
<author>Leon Liangyu Chen, Haoyu Ma, Zhipeng Fan, Ziqi Huang, Animesh Sinha, Xiaoliang Dai, Jialiang Wang, Zecheng He, Jianwei Yang, Chunyuan Li, Junzhe Sun, Chu Wang, Serena Yeung-Levy, Felix Juefei-Xu</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Computer Vision</category>
</item>
<item>
<title>AIRS-Bench: a Suite of Tasks for Frontier AI Research Science Agents</title>
<description>LLM agents hold significant promise for advancing scientific research. To accelerate this progress, we introduce AIRS-Bench (the AI Research Science Benchmark), a suite of 20 tasks sourced from state-of-the-art machine learning papers. These tasks span diverse domains, including language modeling, mathematics, bioinformatics, and time series forecasting. AIRS-Bench tasks assess agentic capabilities over the full research lifecycle---including idea generation, experiment analysis and iterative refinement---without providing baseline code. The AIRS-Bench task format is versatile, enabling easy integration of new tasks and rigorous comparison across different agentic frameworks. We establish baselines using frontier models paired with both sequential and parallel scaffolds. Our results show that agents exceed human SOTA in four tasks but fail to match it in sixteen others. Even when agents surpass human benchmarks, they do not reach the theoretical performance ceiling for the underlying tasks. These findings indicate that AIRS-Bench is far from saturated and offers substantial room for improvement.
We open-source the AIRS-Bench task definitions and evaluation code to catalyze further development in autonomous scientific research.</description>
<link>https://ai.meta.com/research/publications/airs-bench-a-suite-of-tasks-for-frontier-ai-research-science-agents/</link>
<guid isPermaLink="false">2685677328474642</guid>
<pubDate>Tue, 10 Feb 2026 08:00:00 GMT</pubDate>
<author>Alisia Lupidi, Bhavul Gauri, Thomas Simon Foster, Bassel Al Omari, Despoina Magka, Alberto Pepe, Alexis Audran-Reiss, Muna Aghamelu, Nicolas Baldwin, Lucia Cipolina-Kun, Jean-Christophe Gagnon-Audet, Chee Hau Leow, Sandra Lefdal, Hossam Mossalam, Abhinav Moudgil, Saba Nazir, Emanuel Tewolde, Isabel Urrego, Jordi Armengol-Estape, Amar Budhiraja, Gaurav Chaurasia, Abhishek Charnalia, Derek Dunfield, Karen Hambardzumyan, Daniel Izcovich, Martin Josifoski, Ishita Mediratta, Kelvin Niu, Parth Pathak, Michael Shvartsman, Edan Toledo, Anton Protopopov, Roberta Raileanu, Alexander Miller, Tatiana Shavrina, Jakob Foerster, Yoram Bachrach</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0UGM9eu1JmrXKYJlUVM6uXVPhOYc5k56rQOoflJwNZAg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
</item>
<item>
<title>Reducing Government Costs and Increasing Access to Greenspaces in the United Kingdom with DINO</title>
<description>Meta's DINOv2 model is enhancing reforestation efforts around the world. Learn how the UK government is using DINO to help reduce costs and increase access to greenspaces.</description>
<link>https://ai.meta.com/blog/forest-research-dino/</link>
<guid isPermaLink="false">1369350751606207</guid>
<pubDate>Mon, 09 Feb 2026 17:25:45 GMT</pubDate>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.2365-6/631673937_4911145895833160_8225979506389394021_n.gif?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=N94hGWEZkvUQ7kNvwFjZ8Yo&_nc_oc=Adqc7mnraWiwtLCGQIcOu8yhGrf3BlxZ7tS80eVhUCVH2KjTnoQj7hIkwIYd5VoJi50&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af35FdwrXiuXCMDr4wTMR2e93QCGvX63hPrT3F_hySWaQQ&oe=6A08CAFF" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
<category>Open Source</category>
</item>
<item>
<title>Multi-room Apartments Simulation (MRAS) Dataset | Meta AI Research</title>
<description>The Multi-Room Apartments Simulation (MRAS) dataset is a multi-modal dataset created for the task of estimating spatially-distributed acoustic parameters in complex scenes.</description>
<link>https://ai.meta.com/datasets/mras-simulated-rooms-dataset/</link>
<guid isPermaLink="false">2247220155683256</guid>
<pubDate>Wed, 28 Jan 2026 18:49:18 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/612410014_2220471548441535_606775622330495537_n.jpg?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=1bE1l45d6ZQQ7kNvwGzExIi&_nc_oc=Ado-6Ct9Xx3eWjk90FaKNEtJode5O_FsHbdvIUqlBg0pmSXdVV101WYNtSFvKg9wQLc&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af3qXB-t38V8R7cDA3d2X_ydalmNulPb5GudV2l_xyobgw&oe=6A08CB19" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
<item>
<title>How DINO and SAM are Helping Modernize Essential Medical Triage Practices</title>
<description>By leveraging advanced AI models, teams at the University of Pennsylvania are aiming to bring cutting-edge automation to emergency response.</description>
<link>https://ai.meta.com/blog/upenn-dino-sam-helping-medical-triage/</link>
<guid isPermaLink="false">3255814814579669</guid>
<pubDate>Thu, 18 Dec 2025 18:00:20 GMT</pubDate>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.2365-6/597680380_857278207264023_6561532616114241333_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=2MKpeQyaIXMQ7kNvwEEo7tY&_nc_oc=Adpq6zSTC_WtSMZ--D1-Nop-5rrWQy13gjmc5RtnZOIiPEsKiQnSqcnrgEWUNyFyi-0&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0soSTzL3aeIX1NKzi1_kBdrQproswQ_G_Yj1YtbL0imQ&oe=6A08D437" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>The Universities Space Research Association Applies Segment Anything Model for Responding to Flood Emergencies</title>
<description>The Universities Space Research Association and Meta are collaborating to help support water observing systems set up by the U.S. Geological Survey. </description>
<link>https://ai.meta.com/blog/usra-sam-flood-emergencies/</link>
<guid isPermaLink="false">1381613153671193</guid>
<pubDate>Thu, 18 Dec 2025 18:00:12 GMT</pubDate>
<enclosure url="https://scontent-hkg4-2.xx.fbcdn.net/v/t39.2365-6/600304853_1519099669397813_6606203512792368025_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=pBY1dQ-wuD4Q7kNvwEWzj6M&_nc_oc=AdpdwPfP0JGOaTgweBs95cQDXBToqdvNvtjoFQL4YZIyjiTJR4XYN5Ob9gFskGyfO-E&_nc_zt=14&_nc_ht=scontent-hkg4-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af1riT4ErYg3B5mBhneFQSCHg6zIRLwBK7mfOatwMuAbpA&oe=6A08B857" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>How Orakl Oncology is using DINOv2 to accelerate cancer treatment discovery</title>
<description>Orakl Oncology aims to accelerate cancer research and drug development by combining experimental, lab-based insights and machine learning.</description>
<link>https://ai.meta.com/blog/orakl-oncology-dinov2-accelerating-cancer-treatment/</link>
<guid isPermaLink="false">1035491295071881</guid>
<pubDate>Thu, 20 Feb 2025 17:00:31 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/480460718_649637170770478_3994719197111548806_n.jpg?_nc_cat=106&ccb=1-7&_nc_sid=e280be&_nc_ohc=FE_ybvE6IoMQ7kNvwHPNCAd&_nc_oc=Adr7EI6QaBcpn_YJLf8HybvAzlgpRsVmByc2YK5WgnzN5icjGiL0HXeoQ6zPio2ABVw&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af1hqjIF_DtCHIcQURF5zQ7fp7av54qEcmGcOVrHA7NGmA&oe=6A08C327" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
<item>
<title>Aaron Defazio</title>
<description>Aaron's research focuses on improving the practice of machine learning through the development of more reliable and theoretically sound methods such as ...</description>
<link>https://ai.meta.com/people/1115638629589333/aaron-defazio/</link>
<guid isPermaLink="false">1115638629589333</guid>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.35477-6/431543048_2088159024878120_1553708304445833368_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=109&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=vd7kajaxkn4Q7kNvwEG4SoE&_nc_oc=Adrjo9WKzGduyHihaJ9oRKDhf_u31CRkHOA0HsecIfZO9KzGgWUCVqREJtVAKwUQ7Z8&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af2dL0VIYjq62ow8eyYS8LjPVxvX27uZHW5l8cBZ9jYmdw&oe=69F449C8" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Speech & Audio</category>
<category>NLP</category>
<category>Core Machine Learning</category>
<category>Theory</category>
</item>
<item>
<title>Abhimanyu Dubey</title>
<description>Abhimanyu is a Research Scientist at Facebook AI Research working on problems in machine learning. He received his Ph.D. in 2021 from the Massachusetts ...</description>
<link>https://ai.meta.com/people/1401136157273652/abhimanyu-dubey/</link>
<guid isPermaLink="false">1401136157273652</guid>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.35477-6/433945609_1344738226210370_2541979406521528116_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=54biatSobnEQ7kNvwG8A8ef&_nc_oc=AdpIQMfdVEP3_j1Uz5oXdMdPzutCxBnBHAZGm-YnIZrdUDNjyYZHYzJihUGycEfbl7U&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af3L3PyicRwBkD4z1swMg1HkaEdRwunau3pYNMeWnX7_PQ&oe=69F4610D" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Computer Vision</category>
<category>Human & Machine Intelligence</category>
<category>Core Machine Learning</category>
<category>Research</category>
<category>ML Applications</category>
</item>
<item>
<title>Abhishek Das</title>
<description>Abhishek is a Research Scientist at Facebook AI Research (FAIR). His research focuses on deep learning and its applications in climate change, and in bu...</description>
<link>https://ai.meta.com/people/401335699167238/abhishek-das/</link>
<guid isPermaLink="false">401335699167238</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431537252_1543342826243118_6010155505132587328_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=102&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=7bKy_GgNvJUQ7kNvwF8DKlL&_nc_oc=AdoEjkQpXT8Ra71Iwlr5ftbfjSjVC-TbEkqvvy2zTC8X5euaGIYkMJEJO9oLdq0HAzc&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af2TIGpqfxwu5hiFqkul6H7ezW9mjjIknJI274KmSwPCxw&oe=69F44FDB" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
<category>Research</category>
</item>
<item>
<title>Adam Polyak</title>
<description>Adam is a research engineer at Facebook AI Research (FAIR), focusing on deep learning, cross domain image generation, audio generation, and voice conver...</description>
<link>https://ai.meta.com/people/1434021433901222/adam-polyak/</link>
<guid isPermaLink="false">1434021433901222</guid>
<enclosure url="https://scontent-hkg4-2.xx.fbcdn.net/v/t39.35477-6/431518452_1124340545357794_1947310908456905622_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=111&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=qnsH_8koyHsQ7kNvwGzcNSM&_nc_oc=AdoJwpP9r3yeYpvkAnn-ViHaVGRWRZrNmZdEy7rSFN1ArYwY8oe5FRSvKafs-qATreY&_nc_zt=14&_nc_ht=scontent-hkg4-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af18T3LQ2wA7AodseafCTOItHm6LIGTUIxmhKrNymF4iFg&oe=69F46FBA" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Research</category>
<category>ML Applications</category>
<category>Speech & Audio</category>
</item>
<item>
<title>Adeel Cheema</title>
<description>Adeel is a software engineer working on transparency &amp; control within Meta’s Responsible AI pillar. He focuses on creating product experiences to expl...</description>
<link>https://ai.meta.com/people/1808244339639715/adeel-cheema/</link>
<guid isPermaLink="false">1808244339639715</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431551867_767017845524520_8549663744281985032_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=3dHbSn4toEQQ7kNvwF1n4d-&_nc_oc=AdqXznOtgkf_joyKpadW76WIivj8knZ7YKnKZN7RH7Q1e9XhZ_5ak48_CyT-D43hes8&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af1Oddi2UE-zXQg7ao9cUETW6lR2o4luN7rzK8yWvGcVYA&oe=69F44D79" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Adina Williams</title>
<description>Adina is a Research Scientist at Facebook AI Research in NYC (started October 2018). Previously, she earned her PhD at New York University in the Depart...</description>
<link>https://ai.meta.com/people/1396973444287406/adina-williams/</link>
<guid isPermaLink="false">1396973444287406</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431521840_2635057386653799_4484141047373535802_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=104&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=D9tP58vAF5kQ7kNvwF7C6fw&_nc_oc=AdpvPt7C_Jjhp4joD3b5UB-POrJZepNkHFJ0BaE0I8T4EVB_gP9eS-VRbIDEzbFKXCs&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af3aK8L2bXNxmBMVK1pygz7r6-7uQTDuKzJFDTErCAzSDA&oe=69F467EC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Computer Vision</category>
<category>Core Machine Learning</category>
<category>Open Source</category>
<category>NLP</category>
<category>Conversational AI</category>
<category>Human & Machine Intelligence</category>
<category>ML Applications</category>
</item>
<item>
<title>Adriana Romero Soriano</title>
<description>Adriana is currently a research scientist at Facebook AI Research and an adjunct professor at McGill University. The goal of her research is to develop ...</description>
<link>https://ai.meta.com/people/425710613281765/adriana-romero-soriano/</link>
<guid isPermaLink="false">425710613281765</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/431577893_1361257101250742_6288541672250118710_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=nsGSoSWY8rkQ7kNvwHp-g62&_nc_oc=AdpqSlSYXprfSM2Tg780n64EZrGd82vlPOVyEfr8PRGp2ppWUVoQzwDtVilqFfNf3lI&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af3JltHpeKtIoyM3i3FDIi8_2eymgTNPV_ALzfDwNw7ibg&oe=69F44DCC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Reinforcement Learni9ng</category>
<category>Computer Vision</category>
<category>Research</category>
<category>NLP</category>
<category>Core Machine Learning</category>
<category>Conversational AI</category>
</item>
<item>
<title>Akhil Mathur</title>
<description>Akhil Mathur is a Research Scientist in the Llama Research team at Meta, working on enhancing the reasoning abilities of Llama models. Previously, as a ...</description>
<link>https://ai.meta.com/people/1251205439201018/akhil-mathur/</link>
<guid isPermaLink="false">1251205439201018</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/457008091_844600247808930_1170475198447093629_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=A2xSui7yUHAQ7kNvwFS23zM&_nc_oc=AdrahVtA7Wbc1v3tY3QbRSHzsNwyMLaCTMf2PTn736NWv6hhVgubMa_4-14IsGUrHK4&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af2fYoN8J4xLsEei0JOr6UBHHj8N7MrjN5lJ8Uo2Job__A&oe=69F4499C" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Akshara Rai</title>
<description>Akshara is a Research Scientist at Facebook AI Research, working at the intersection of machine learning and control. Her research aims at teaching robo...</description>
<link>https://ai.meta.com/people/423502423531745/akshara-rai/</link>
<guid isPermaLink="false">423502423531745</guid>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.35477-6/433923962_1381208202581063_6502596142308749692_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=0Wm1dG_MC68Q7kNvwEQP7ar&_nc_oc=Adq_deH7ji9ZB1xmu-dPrpWID0BCX9-6V1N9jAnvZL94SgSLUQTUbTggiNCwTtCsqhY&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0xrIvbGwL3UAx9jtyJaAj0iy2xygtEavTQFXIvG2C6_g&oe=69F43F4F" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Robotics</category>
<category>Computer Vision</category>
<category>Research</category>
<category>Reinforcement Learni9ng</category>
</item>
<item>
<title>Alborz Geramifard</title>
<description>Alborz Geramifard is a senior research manager at Meta AI supporting the Conversational AI. Prior to joining Meta, he led the conversational AI team at ...</description>
<link>https://ai.meta.com/people/363544420007153/alborz-geramifard/</link>
<guid isPermaLink="false">363544420007153</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431495520_3457483597876936_681607123433531609_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=QzBw61zVxHEQ7kNvwFlE51j&_nc_oc=Adq6JhqFdOijHej1827AxfrBteLpqdbt7M2AEiSaSYBRb0twHVO0Vkzc7La_KpTtqrw&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af1DjqTo6y2-LnbmqCkAtYPj7ZD7KRBiESzlywAvrxzEFA&oe=69F45A51" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Conversational AI</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
</item>
<item>
<title>Alexander H. Miller</title>
<description>Alexander has worked in multi-domain research engineering management, supporting research engineers in London, Tel Aviv, and New York. He is advanced in...</description>
<link>https://ai.meta.com/people/264624513377272/alexander-h-miller/</link>
<guid isPermaLink="false">264624513377272</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/433950826_944833516851120_841662693953017766_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=6ftFVCshb-8Q7kNvwEmShVr&_nc_oc=AdqDT05sTcWqiXL8CooTGILItRGi0hr8Ylblc4YGYl0M5hVqjpjk1UwnAs5-tmRvHo8&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af1a1y_fcZ7vBytU4qK5Yo5YDfkAhqTuOQZLnInCmc47sQ&oe=69F45C6C" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Reinforcement Learni9ng</category>
<category>NLP</category>
<category>Graphics</category>
</item>
<item>
<title>Amey Porobo Dharwadker</title>
<description>Amey Porobo Dharwadker works as a Machine Learning Engineering Manager, leading Facebook's Video Recommendations Ranking team. Renowned for his expertis...</description>
<link>https://ai.meta.com/people/393924029921120/amey-porobo-dharwadker/</link>
<guid isPermaLink="false">393924029921120</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/431510720_7490594577724788_4289788014840147933_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=106&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=CDFvi0iDZRcQ7kNvwHngb7S&_nc_oc=Ado-on7H7BL3gsoNI2IvWjApO7MmYjUwGlBV_K3EOEWlY71TKmdUsBi5V96q7WVtg28&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=-Jmb5VLjy4Pb4QUqsuC2kw&_nc_ss=73289&oh=00_Af0f570Z8NjCqcDKA9Mh8Jq5u5Z8I8mB6FF74ekv5iaY8Q&oe=69F440F5" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
</channel>
</rss>... |
Contributor
http://localhost:1200/meta/ai/global-search/content_types=blog - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — content_types=blog</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/content_types=blog" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/. - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:22:26 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>Scaling How We Build and Test Our Most Advanced AI</title>
<description>As we build more capable, personalized AI, reliability, security, and user protections are more important than ever. </description>
<link>https://ai.meta.com/blog/scaling-how-we-build-test-advanced-ai/</link>
<guid isPermaLink="false">4369469056623958</guid>
<pubDate>Wed, 08 Apr 2026 17:00:17 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/668130848_1453339552354697_3291493920555502062_n.png?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=yvFttOBoRYkQ7kNvwHAGoUI&_nc_oc=AdrMTKarV4pYb-C6Lo7cZ4FS8VtfHNic415DSbyvO9pcVHrt8wGk3pW7eGF_3ATLqB8&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af1frodUWoKWOD5ZGzS0EwwFJgArXMummuNyDAK2jrtFpg&oe=6A08CDEE" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Introducing Muse Spark: Scaling Towards Personal Superintelligence</title>
<description></description>
<link>https://ai.meta.com/blog/introducing-muse-spark-msl/</link>
<guid isPermaLink="false">963191949487556</guid>
<pubDate>Wed, 08 Apr 2026 16:00:59 GMT</pubDate>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.2365-6/667507363_1871749950191919_4996140970371485757_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=IbntS1cBAcsQ7kNvwEkZk-q&_nc_oc=Ado4EZgw05h6efdlgHKUmEWqiQAUivkkK5wab1s8qYcxECiinRzqsBcSp7BtuIVOo5k&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af0qyqFeqXW26PYd2VaexFaaDuJHhl5Om8tTBQbLB6uFPQ&oe=6A08D22C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>How Alta Daily Uses Meta’s Segment Anything to Reimagine the Digital Closet</title>
<description></description>
<link>https://ai.meta.com/blog/alta-daily-fashion-app-segment-anything/</link>
<guid isPermaLink="false">2194788291268065</guid>
<pubDate>Mon, 06 Apr 2026 18:53:27 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/658967078_993415586351079_325919381939729048_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=cBKF2Wk0XYoQ7kNvwGT68WW&_nc_oc=Adr_OUg82eUw0EbluNihH00lRvwhw0cI82roesUhcZVs9_YWPOnE24LdOGDODhZ8PXE&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af3TlYfE02OjtoXRMuFSXZOm-C9Z-G2mGCE1hFwNjdHcaQ&oe=6A08B252" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning</title>
<description></description>
<link>https://ai.meta.com/blog/segment-anything-model-3/</link>
<guid isPermaLink="false">715047027731045</guid>
<pubDate>Fri, 27 Mar 2026 16:00:37 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/583339165_852765810541557_5592130597432767690_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=WvtkkpGzh44Q7kNvwHtq_BU&_nc_oc=AdqOSUbFE1HjLhxvY58CCZUsdmat-y6v9A-cIg_tkySMD-A2NT_eE5pWvyzi6Gtrlq0&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af1tApWwSTT-fRmYvqTiWOrVfqkgWCWC7yklz5BAAAzKhw&oe=6A08D061" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>Introducing TRIBE v2: A Predictive Foundation Model Trained to Understand How the Human Brain Processes Complex Stimuli</title>
<description></description>
<link>https://ai.meta.com/blog/tribe-v2-brain-predictive-foundation-model/</link>
<guid isPermaLink="false">25512119895129580</guid>
<pubDate>Thu, 26 Mar 2026 16:00:16 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/659112495_952140057346056_1728296311373871062_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=rBe_VZxcFD4Q7kNvwEl0TGK&_nc_oc=Adq6bzVjbryLAhis8bGJPU1Umwebg8J3Ul_Sn-xoSPOJfZTVx6fNEnrsqhxPP2J05wc&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af3FpngNRtHkE7KZXdBRajSzkGFUzBSG0mn1J1psG6aZAg&oe=6A08AF6B" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Research</category>
</item>
<item>
<title>Four MTIA Chips in Two Years: Scaling AI Experiences for Billions</title>
<description>Serving a wide range of AI models on a global scale, while maintaining the lowest possible costs, is one of the most demanding infrastructure challenges in the industry. </description>
<link>https://ai.meta.com/blog/meta-mtia-scale-ai-chips-for-billions/</link>
<guid isPermaLink="false">814227801000565</guid>
<pubDate>Wed, 11 Mar 2026 14:00:55 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/649719916_2147323009439466_6866394022243367245_n.png?_nc_cat=103&ccb=1-7&_nc_sid=e280be&_nc_ohc=HQUtIT-12s8Q7kNvwGOLZwg&_nc_oc=AdrxxwjcxH_agF9UMFF-AQWK2vy12eCF5p5jf1ozY-qI61a_wgoP7vvJizTyzRxV0xc&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af2Z1H2uWkrsXg448oUv-7uGfnAygE_ziBNhhqgHzgu3zw&oe=6A08CB6C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Mapping the World's Forests with Greater Precision: Introducing Canopy Height Maps v2</title>
<description>In partnership with the World Resources Institute, today we’re announcing Canopy Height Maps v2 (CHMv2), an open source model, along with world-scale maps generated with the model.</description>
<link>https://ai.meta.com/blog/world-resources-institute-dino-canopy-height-maps-v2/</link>
<guid isPermaLink="false">1675459620286410</guid>
<pubDate>Tue, 10 Mar 2026 16:00:06 GMT</pubDate>
<enclosure url="https://scontent-hkg4-2.xx.fbcdn.net/v/t39.2365-6/644822799_1500169394861918_3213642176014981812_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=5tvZOejkl8MQ7kNvwE3rzOr&_nc_oc=AdpUWj25o490QP_XxwEWiK2CqMWOSxEQWHDJ-K-g7z3HzWbeYKo0AeH72yUz_8SC2Pk&_nc_zt=14&_nc_ht=scontent-hkg4-2.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af1SJBgCSVssX0d-JEjrsMsyLzi-EBVM82uybpKcn3MFDg&oe=6A08D606" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
<item>
<title>Reducing Government Costs and Increasing Access to Greenspaces in the United Kingdom with DINO</title>
<description>Meta's DINOv2 model is enhancing reforestation efforts around the world. Learn how the UK government is using DINO to help reduce costs and increase access to greenspaces.</description>
<link>https://ai.meta.com/blog/forest-research-dino/</link>
<guid isPermaLink="false">1369350751606207</guid>
<pubDate>Mon, 09 Feb 2026 17:25:45 GMT</pubDate>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.2365-6/631673937_4911145895833160_8225979506389394021_n.gif?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=N94hGWEZkvUQ7kNvwFjZ8Yo&_nc_oc=Adqc7mnraWiwtLCGQIcOu8yhGrf3BlxZ7tS80eVhUCVH2KjTnoQj7hIkwIYd5VoJi50&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af1WZ4yzIzmYVlROQPRmUqfvxNOfhu4IafgkzsILJB3A2w&oe=6A08CAFF" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
<category>Open Source</category>
</item>
<item>
<title>How DINO and SAM are Helping Modernize Essential Medical Triage Practices</title>
<description>By leveraging advanced AI models, teams at the University of Pennsylvania are aiming to bring cutting-edge automation to emergency response.</description>
<link>https://ai.meta.com/blog/upenn-dino-sam-helping-medical-triage/</link>
<guid isPermaLink="false">3255814814579669</guid>
<pubDate>Thu, 18 Dec 2025 18:00:20 GMT</pubDate>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.2365-6/597680380_857278207264023_6561532616114241333_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=2MKpeQyaIXMQ7kNvwEEo7tY&_nc_oc=Adpq6zSTC_WtSMZ--D1-Nop-5rrWQy13gjmc5RtnZOIiPEsKiQnSqcnrgEWUNyFyi-0&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af3wxJste55coV6TPSZ-3y8_-lQi0tdNp46YaRoyTv0mLA&oe=6A08D437" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>The Universities Space Research Association Applies Segment Anything Model for Responding to Flood Emergencies</title>
<description>The Universities Space Research Association and Meta are collaborating to help support water observing systems set up by the U.S. Geological Survey. </description>
<link>https://ai.meta.com/blog/usra-sam-flood-emergencies/</link>
<guid isPermaLink="false">1381613153671193</guid>
<pubDate>Thu, 18 Dec 2025 18:00:12 GMT</pubDate>
<enclosure url="https://scontent-hkg4-2.xx.fbcdn.net/v/t39.2365-6/600304853_1519099669397813_6606203512792368025_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=pBY1dQ-wuD4Q7kNvwEWzj6M&_nc_oc=AdpdwPfP0JGOaTgweBs95cQDXBToqdvNvtjoFQL4YZIyjiTJR4XYN5Ob9gFskGyfO-E&_nc_zt=14&_nc_ht=scontent-hkg4-2.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af2BHzagStTzeb6KkqCsVDepCXrqYAk1PjXduAGtw3YO2Q&oe=6A08B857" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>Introducing SAM Audio: The First Unified Multimodal Model for Audio Separation</title>
<description>SAM Audio transforms audio processing by making it easy to isolate any sound from complex audio mixtures using natural, multimodal prompts — whether through text, visual cues, or marking time segments. </description>
<link>https://ai.meta.com/blog/sam-audio/</link>
<guid isPermaLink="false">1658975421738456</guid>
<pubDate>Tue, 16 Dec 2025 18:00:17 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/600517835_858099609958067_2071268735982577268_n.png?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=jNAx0Vnb61IQ7kNvwFwRqqt&_nc_oc=Adr1BW16QLNYFB5SUPFPJ3dLCZFcBgYlkvlH6A4afN06QWlE5vcGNnfNLqAllQRpGhE&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af1YA6ENj7DKtxircnfnQZOowxs0ZYhn0G0kh0zVxv4azg&oe=6A08B61B" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>How Orakl Oncology is using DINOv2 to accelerate cancer treatment discovery</title>
<description>Orakl Oncology aims to accelerate cancer research and drug development by combining experimental, lab-based insights and machine learning.</description>
<link>https://ai.meta.com/blog/orakl-oncology-dinov2-accelerating-cancer-treatment/</link>
<guid isPermaLink="false">1035491295071881</guid>
<pubDate>Thu, 20 Feb 2025 17:00:31 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/480460718_649637170770478_3994719197111548806_n.jpg?_nc_cat=106&ccb=1-7&_nc_sid=e280be&_nc_ohc=FE_ybvE6IoMQ7kNvwHPNCAd&_nc_oc=Adr7EI6QaBcpn_YJLf8HybvAzlgpRsVmByc2YK5WgnzN5icjGiL0HXeoQ6zPio2ABVw&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=mDSfWgTSmqSVtIMNHeN93Q&_nc_ss=73289&oh=00_Af0jgyB-3hK0BSx117K8QjYpFPG4hYWqK7qYIjdud80fMA&oe=6A08C327" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
</channel>
</rss>http://localhost:1200/meta/ai/global-search/content_types=person - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — content_types=person</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/content_types=person" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/. - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:22:30 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>Aaron Defazio</title>
<description>Aaron's research focuses on improving the practice of machine learning through the development of more reliable and theoretically sound methods such as ...</description>
<link>https://ai.meta.com/people/1115638629589333/aaron-defazio/</link>
<guid isPermaLink="false">1115638629589333</guid>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.35477-6/431543048_2088159024878120_1553708304445833368_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=109&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=vd7kajaxkn4Q7kNvwEG4SoE&_nc_oc=Adrjo9WKzGduyHihaJ9oRKDhf_u31CRkHOA0HsecIfZO9KzGgWUCVqREJtVAKwUQ7Z8&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af2sBaTyE905asyOGznyoaoJAu9d2yY4WjUDzo9bEIyO-w&oe=69F449C8" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Speech & Audio</category>
<category>NLP</category>
<category>Core Machine Learning</category>
<category>Theory</category>
</item>
<item>
<title>Abhimanyu Dubey</title>
<description>Abhimanyu is a Research Scientist at Facebook AI Research working on problems in machine learning. He received his Ph.D. in 2021 from the Massachusetts ...</description>
<link>https://ai.meta.com/people/1401136157273652/abhimanyu-dubey/</link>
<guid isPermaLink="false">1401136157273652</guid>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.35477-6/433945609_1344738226210370_2541979406521528116_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=54biatSobnEQ7kNvwG8A8ef&_nc_oc=AdpIQMfdVEP3_j1Uz5oXdMdPzutCxBnBHAZGm-YnIZrdUDNjyYZHYzJihUGycEfbl7U&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af3BcvQ_GC8AXJZifoSKjAVLwYDxN5olR3XzXfnG2O0Tgg&oe=69F4610D" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Computer Vision</category>
<category>Research</category>
<category>ML Applications</category>
<category>Human & Machine Intelligence</category>
<category>Core Machine Learning</category>
</item>
<item>
<title>Abhishek Das</title>
<description>Abhishek is a Research Scientist at Facebook AI Research (FAIR). His research focuses on deep learning and its applications in climate change, and in bu...</description>
<link>https://ai.meta.com/people/401335699167238/abhishek-das/</link>
<guid isPermaLink="false">401335699167238</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431537252_1543342826243118_6010155505132587328_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=102&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=7bKy_GgNvJUQ7kNvwF8DKlL&_nc_oc=AdoEjkQpXT8Ra71Iwlr5ftbfjSjVC-TbEkqvvy2zTC8X5euaGIYkMJEJO9oLdq0HAzc&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af0cE427wx2fil3Ve1nHggQL2tD5P-bDP593fDSSf2at3A&oe=69F44FDB" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Computer Vision</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
<category>NLP</category>
</item>
<item>
<title>Adam Polyak</title>
<description>Adam is a research engineer at Facebook AI Research (FAIR), focusing on deep learning, cross domain image generation, audio generation, and voice conver...</description>
<link>https://ai.meta.com/people/1434021433901222/adam-polyak/</link>
<guid isPermaLink="false">1434021433901222</guid>
<enclosure url="https://scontent-hkg4-2.xx.fbcdn.net/v/t39.35477-6/431518452_1124340545357794_1947310908456905622_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=111&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=qnsH_8koyHsQ7kNvwGzcNSM&_nc_oc=AdoJwpP9r3yeYpvkAnn-ViHaVGRWRZrNmZdEy7rSFN1ArYwY8oe5FRSvKafs-qATreY&_nc_zt=14&_nc_ht=scontent-hkg4-2.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af18i8UKJChhZHjDF6LtIyI4YDVh3mJ-rCQww17UAwTZoA&oe=69F46FBA" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Research</category>
<category>ML Applications</category>
<category>Speech & Audio</category>
</item>
<item>
<title>Adeel Cheema</title>
<description>Adeel is a software engineer working on transparency &amp; control within Meta’s Responsible AI pillar. He focuses on creating product experiences to expl...</description>
<link>https://ai.meta.com/people/1808244339639715/adeel-cheema/</link>
<guid isPermaLink="false">1808244339639715</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431551867_767017845524520_8549663744281985032_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=3dHbSn4toEQQ7kNvwF1n4d-&_nc_oc=AdqXznOtgkf_joyKpadW76WIivj8knZ7YKnKZN7RH7Q1e9XhZ_5ak48_CyT-D43hes8&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af3TGDhorh7aUPoAMdTqswp8pgI0gx9JnPgYRveTDDgjxg&oe=69F44D79" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Adina Williams</title>
<description>Adina is a Research Scientist at Facebook AI Research in NYC (started October 2018). Previously, she earned her PhD at New York University in the Depart...</description>
<link>https://ai.meta.com/people/1396973444287406/adina-williams/</link>
<guid isPermaLink="false">1396973444287406</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431521840_2635057386653799_4484141047373535802_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=104&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=D9tP58vAF5kQ7kNvwF7C6fw&_nc_oc=AdpvPt7C_Jjhp4joD3b5UB-POrJZepNkHFJ0BaE0I8T4EVB_gP9eS-VRbIDEzbFKXCs&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af0J0eVvX2div_zWGxkIcOVmKk_hk6vQ0kmnDNyOT2imsQ&oe=69F467EC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Computer Vision</category>
<category>Core Machine Learning</category>
<category>Open Source</category>
<category>NLP</category>
<category>Conversational AI</category>
<category>Human & Machine Intelligence</category>
<category>ML Applications</category>
</item>
<item>
<title>Adriana Romero Soriano</title>
<description>Adriana is currently a research scientist at Facebook AI Research and an adjunct professor at McGill University. The goal of her research is to develop ...</description>
<link>https://ai.meta.com/people/425710613281765/adriana-romero-soriano/</link>
<guid isPermaLink="false">425710613281765</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/431577893_1361257101250742_6288541672250118710_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=nsGSoSWY8rkQ7kNvwHp-g62&_nc_oc=AdpqSlSYXprfSM2Tg780n64EZrGd82vlPOVyEfr8PRGp2ppWUVoQzwDtVilqFfNf3lI&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af3e6bTtVCGg21d7fugsyXIIW9sv-yySTgShnjuxxOtvyw&oe=69F44DCC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Core Machine Learning</category>
<category>Conversational AI</category>
<category>Reinforcement Learni9ng</category>
</item>
<item>
<title>Akhil Mathur</title>
<description>Akhil Mathur is a Research Scientist in the Llama Research team at Meta, working on enhancing the reasoning abilities of Llama models. Previously, as a ...</description>
<link>https://ai.meta.com/people/1251205439201018/akhil-mathur/</link>
<guid isPermaLink="false">1251205439201018</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/457008091_844600247808930_1170475198447093629_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=A2xSui7yUHAQ7kNvwFS23zM&_nc_oc=AdrahVtA7Wbc1v3tY3QbRSHzsNwyMLaCTMf2PTn736NWv6hhVgubMa_4-14IsGUrHK4&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af391yQV1Rf7LFe9-PrqloMmapDqEIBsREKA42aua2EqYw&oe=69F4499C" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Akshara Rai</title>
<description>Akshara is a Research Scientist at Facebook AI Research, working at the intersection of machine learning and control. Her research aims at teaching robo...</description>
<link>https://ai.meta.com/people/423502423531745/akshara-rai/</link>
<guid isPermaLink="false">423502423531745</guid>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.35477-6/433923962_1381208202581063_6502596142308749692_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=0Wm1dG_MC68Q7kNvwEQP7ar&_nc_oc=Adq_deH7ji9ZB1xmu-dPrpWID0BCX9-6V1N9jAnvZL94SgSLUQTUbTggiNCwTtCsqhY&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af1OEFeZy77ZhmpwI0jbYexei8_iswcbwIrWbyLvRvPZ7g&oe=69F43F4F" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Robotics</category>
<category>Computer Vision</category>
<category>Reinforcement Learni9ng</category>
<category>Research</category>
</item>
<item>
<title>Alborz Geramifard</title>
<description>Alborz Geramifard is a senior research manager at Meta AI supporting the Conversational AI. Prior to joining Meta, he led the conversational AI team at ...</description>
<link>https://ai.meta.com/people/363544420007153/alborz-geramifard/</link>
<guid isPermaLink="false">363544420007153</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431495520_3457483597876936_681607123433531609_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=QzBw61zVxHEQ7kNvwFlE51j&_nc_oc=Adq6JhqFdOijHej1827AxfrBteLpqdbt7M2AEiSaSYBRb0twHVO0Vkzc7La_KpTtqrw&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af15sS6BqHwzGsnmdW8YvimAXYi__3VoPXB82nuVa48_6w&oe=69F45A51" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
<category>Conversational AI</category>
<category>NLP</category>
<category>Computer Vision</category>
</item>
<item>
<title>Alexander H. Miller</title>
<description>Alexander has worked in multi-domain research engineering management, supporting research engineers in London, Tel Aviv, and New York. He is advanced in...</description>
<link>https://ai.meta.com/people/264624513377272/alexander-h-miller/</link>
<guid isPermaLink="false">264624513377272</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/433950826_944833516851120_841662693953017766_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=6ftFVCshb-8Q7kNvwEmShVr&_nc_oc=AdqDT05sTcWqiXL8CooTGILItRGi0hr8Ylblc4YGYl0M5hVqjpjk1UwnAs5-tmRvHo8&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af1Wy4K32JJlB_6W28-rxhGvFsj8FBBafbNBv7W7RY6Wog&oe=69F45C6C" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Reinforcement Learni9ng</category>
<category>NLP</category>
<category>Graphics</category>
</item>
<item>
<title>Amey Porobo Dharwadker</title>
<description>Amey Porobo Dharwadker works as a Machine Learning Engineering Manager, leading Facebook's Video Recommendations Ranking team. Renowned for his expertis...</description>
<link>https://ai.meta.com/people/393924029921120/amey-porobo-dharwadker/</link>
<guid isPermaLink="false">393924029921120</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/431510720_7490594577724788_4289788014840147933_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=106&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=CDFvi0iDZRcQ7kNvwHngb7S&_nc_oc=Ado-on7H7BL3gsoNI2IvWjApO7MmYjUwGlBV_K3EOEWlY71TKmdUsBi5V96q7WVtg28&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=lwdjKAHUukt4SV627Vdj2A&_nc_ss=73289&oh=00_Af1AWtUoXXMU6ppwmRxQbrkXMBqVhS-TCauXW4V6GrgO7w&oe=69F440F5" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
</channel>
</rss>... |
Contributor
http://localhost:1200/meta/ai/global-search/content_types=publication - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — content_types=publication</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/content_types=publication" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/. - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:22:35 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>AIRA₂: Overcoming Bottlenecks in AI Research Agents</title>
<description>Existing research has identified three structural performance bottlenecks in AI research agents: (1) synchronous single-GPU execution constrains sample throughput, limiting the benefit of search; (2) a generalization gap where validation-based selection causes overfitting and performance to degrade over extended search horizons; and (3) the limited capability of fixed, single-turn LLM operators imposes a ceiling on search performance. We introduce AIRA₂, which addresses these bottlenecks through three architectural choices: an asynchronous multi-GPU worker pool that increases experiment throughput linearly; a Hidden Consistent Evaluation protocol that delivers a reliable evaluation signal; and ReAct agents that dynamically scope their actions and debug interactively. On MLE-bench-30, AIRA₂ achieves a mean Percentile Rank of 81.5% at 24 hours and 83.1% at 72 hours, outperforming the strongest baseline, which achieves 72.7%. On AIRS-Bench, AIRA₂ exceeds human state-of-the-art on 6 out of 20 diverse research tasks. Ablations confirm that each architectural component is necessary, that performance follows a predictable scaling law that transfers across LLM backbones, and that the "overfitting" reported in prior work was driven by evaluation noise rather than true data memorization.</description>
<link>https://ai.meta.com/research/publications/aira-overcoming-bottlenecks-in-ai-research-agents/</link>
<guid isPermaLink="false">1298625375513718</guid>
<pubDate>Thu, 16 Apr 2026 07:00:00 GMT</pubDate>
<author>Karen Hambardzumyan, Nicolas Baldwin, Edan Toledo, Rishi Hazra, Michael Kuchnik, Bassel Al Omari, Thomas Simon Foster, Anton Protopopov, Jean-Christophe Gagnon-Audet, Ishita Mediratta, Kelvin Niu, Michael Shvartsman, Alisia Lupidi, Alexis Audran-Reiss, Parth Pathak, Tatiana Shavrina, Despoina Magka, Hela Momand, Derek Dunfield, Nicola Cancedda, Pontus Stenetorp, Carole-Jean Wu, Jakob Foerster, Yoram Bachrach, Martin Josifoski</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
</item>
<item>
<title>TransText: Transparency Aware Image-to-Video Typography Animation</title>
<description>We introduce the first method, to the best of our knowledge, for adapting image-to-video models to layer-aware text (glyph) animation, a capability critical for practical dynamic visual design. Existing approaches predominantly handle the transparency-encoding (alpha channel) as an extra latent dimension appended to the RGB space, necessitating the reconstruction of the underlying RGB-centric variational autoencoder (VAE). However, given the scarcity of high-quality transparent glyph data, retraining the VAE is computationally expensive and may erode the robust semantic priors learned from massive RGB corpora, potentially leading to latent pattern mixing. To mitigate these limitations, we propose TransText, a framework based on a novel Alpha-as-RGB paradigm to jointly model appearance and transparency without modifying the pre-trained generative manifold. TransText embeds the alpha channel as an RGB-compatible visual signal through latent spatial concatenation, explicitly ensuring strict cross-modal (RGB-and-Alpha) consistency while preventing feature entanglement. Our experiments demonstrate that TransText significantly outperforms baselines, generating coherent, high-fidelity transparent animations with diverse, fine-grained effects.</description>
<link>https://ai.meta.com/research/publications/transtext-transparency-aware-image-to-video-typography-animation/</link>
<guid isPermaLink="false">2172593836888174</guid>
<pubDate>Tue, 14 Apr 2026 07:00:00 GMT</pubDate>
<author>Fei Zhang, Zijian Zhou, Bohao Tang, Sen He, Hang Li (BizAI), Zhe Wang, Soubhik Sanyal, Pengfei Liu, Viktar Atliha, Tao Xiang, Frost Xu, Semih Gunel</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Computer Vision</category>
<category>ML Applications</category>
</item>
<item>
<title>Think in Strokes, Not Pixels: Process-Driven Image Generation via Interleaved Reasoning</title>
<description>Humans paint images incrementally: they plan a global layout, sketch a coarse draft, inspect, and refine details, and most importantly, each step is grounded in the evolving visual states. However, can unified multimodal models trained on text-image interleaved datasets also imagine the chain of intermediate states? In this paper, we introduce process-driven image generation, a multi-step paradigm that decomposes synthesis into an interleaved reasoning trajectory of thoughts and actions. Rather than generating images in a single step, our approach unfolds across multiple iterations, each consisting of 4 stages: textual planning, visual drafting, textual reflection, and visual refinement. The textual reasoning explicitly conditions how the visual state should evolve, while the generated visual intermediate in turn constrains and grounds the next round of textual reasoning. A core challenge of process-driven generation stems from the ambiguity of intermediate states: how can models evaluate each partially-complete image? We address this through dense, step-wise supervision that maintains two complementary constraints: for the visual intermediate states, we enforce the spatial and semantic consistency; for the textual intermediate states, we preserve the prior visual knowledge while enabling the model to identify and correct prompt-violating elements. This makes the generation process explicit, interpretable, and directly supervisable. To validate proposed method, we conduct experiments under various text-to-image generation benchmarks.</description>
<link>https://ai.meta.com/research/publications/think-in-strokes-not-pixels-process-driven-image-generation-via-interleaved-reasoning/</link>
<guid isPermaLink="false">847030005080644</guid>
<pubDate>Thu, 09 Apr 2026 07:00:00 GMT</pubDate>
<author>Lei Zhang, Junjiao Tian, Zhipeng Fan, Kunpeng Li, Jialiang Wang, Weifeng Chen, Markos Georgopoulos, Felix Xu, Yuxiao Bao, Julian McAuley, Manling Li, Zecheng He</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Computer Vision</category>
</item>
<item>
<title>A foundation model of vision, audition, and language for in-silico neuroscience</title>
<description>Cognitive neuroscience is fragmented into specialized models, each tailored to specific experimental paradigms, hence preventing a unified model of cognition in the human brain. Here, we introduce TRIBE v2, a tri-modal (video, audio and language) foundation model capable of predicting human brain activity in a variety of naturalistic and experimental conditions. Leveraging a unified dataset of over 1,000 hours of fMRI across 720 subjects, we demonstrate that our model accurately predicts high-resolution brain responses for novel stimuli, tasks and subjects, superseding traditional linear encoding models, delivering several-fold improvements in accuracy. Critically, TRIBE v2 enables in silico experimentation: tested on seminal visual and neuro-linguistic paradigms, it recovers a variety of results established by decades of empirical research. Finally, by extracting interpretable latent features, TRIBE v2 reveals the fine-grained topography of multisensory integration. These results establish artificial intelligence as a unifying framework for exploring the functional organization of the human brain.</description>
<link>https://ai.meta.com/research/publications/a-foundation-model-of-vision-audition-and-language-for-in-silico-neuroscience/</link>
<guid isPermaLink="false">1009839085553251</guid>
<pubDate>Thu, 26 Mar 2026 07:00:00 GMT</pubDate>
<author>Stéphane d'Ascoli, Jérémy Rapin, Yohann Benchetrit, Teon Brooks, Katelyn Begany, Josephine Raugel, Hubert Jacob Banville, Jean Remi King</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
</item>
<item>
<title>HyperAgents</title>
<description>Self-improving AI systems aim to reduce reliance on human engineering by learning to improve their own learning and problem-solving processes. Existing approaches to recursive self-improvement typically rely on fixed, handcrafted meta-level mechanisms, which fundamentally limit how fast such systems can improve. The Darwin Gödel Machine (DGM)(Zhang et al., 2025b) demonstrates that open-ended self-improvement is achievable in coding. Starting from a single coding agent, the DGM repeatedly generates and evaluates self-modified variants, forming a growing archive of stepping stones for future improvement. Because both evaluation and self-modification are coding tasks, gains in coding ability can translate into gains in self-improvement ability. However, this alignment does not generally hold beyond coding domains.
We introduce hyperagents, self-referential agents that integrate a task agent (which solves the target task) and a meta agent (which modifies itself and the task agent) into a single editable program. Crucially, the meta-level modification procedure is itself editable, enabling metacognitive self-modification, improving not only task-solving behavior, but also the mechanism that generates future improvements. We instantiate this framework by extending DGM to create DGM-Hyperagents (DGM-H). By allowing the improvement procedure to evolve, the DGM-H eliminates the assumption of domain-specific alignment between task performance and self-modification skill, and can potentially support self-accelerating progress on any computable task. Across diverse domains (coding, paper review, robotics reward design, and Olympiad-level math-solution grading), the DGM-H improves performance over time and outperforms baselines without self-improvement or open-ended exploration, as well as prior self-improving systems like DGM. We further show that the DGM-H improves the process by which it generates new agents (e.g., persistent memory, performance tracking), and that these meta-level improvements transfer across domains and accumulate across runs. All experiments were conducted with safety precautions (e.g., sandboxing, human oversight). We discuss what safety entails in this setting and the broader implications of self-improving systems. DGM-Hyperagents offer a glimpse of open-ended AI systems that do not merely search for better solutions, but continually improve their search for how to improve.</description>
<link>https://ai.meta.com/research/publications/hyperagents/</link>
<guid isPermaLink="false">1669124537439877</guid>
<pubDate>Tue, 24 Mar 2026 07:00:00 GMT</pubDate>
<author>Jenny Zhang, Bingchen Zhao, Winnie Yang, Jakob Foerster, Sam Devlin, Tatiana Shavrina</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
<category>Open Source</category>
</item>
<item>
<title>Omnilingual MT: Machine Translation for 1,600 Languages</title>
<description>Advances made through No Language Left Behind (NLLB) have demonstrated that high-quality machine translation (MT) scale to 200 languages. Later Large Language Models (LLMs) have been adopted for MT, increasing in quality but not necessarily extending language coverage. Current systems remain constrained by limited coverage and a persistent generation bottleneck: while crosslingual transfer enables models to somehow understand many undersupported languages, they often cannot generate them reliably, leaving most of the world’s 7,000 languages—especially endangered and marginalized ones—outside the reach of modern MT. Early explorations in extreme scaling offered promising proofs of concept but did not yield sustained solutions.
We present Omnilingual Machine Translation (OMT), the first MT system supporting more than 1,600 languages. This scale is enabled by a comprehensive data strategy that integrates large public multilingual corpora with newly created datasets, including manually curated MeDLEY bitext, synthetic backtranslation, and mining, substantially expanding coverage across long-tail languages, domains, and registers. To ensure both reliable and expansive evaluation, we combined standard metrics with a suite of evaluation artifacts: BLASER 3 quality estimation model (reference-free), OmniTOX toxicity classifier, BOUQuET dataset (a newly created, largest-to-date multilingual evaluation collection built from scratch and manually extended across a wide range of linguistic families), and Met-BOUQuET dataset (faithful multilingual quality estimation at scale). We explore two ways of specializing an LLM for machine translation: as a decoder-only model (OMT-LLaMA) or as a module in an encoder–decoder architecture (OMT-NLLB). The former consists of a model built on LLaMA3, with multilingual continual pretraining and retrieval-augmented translation for inference-time adaptation. The latter is a model built on top of a multilingual aligned space (OmniSONAR, itself also based on LLaMA3), and introduces a training methodology that can exploit non-parallel data, allowing us to incorporate the decoder-only continuous pretraining data into the training of an encoder–decoder architecture. Notably, all our 1B to 8B parameter models match or exceed the MT performance of a 70B LLM baseline, revealing a clear specialization advantage and enabling strong translation quality in low-compute settings. Moreover, our evaluation of English-to-1,600 translations further shows that while baseline models can interpret undersupported languages, they frequently fail to generate them with meaningful fidelity; OMT-LLaMA models substantially expand the set of languages for which coherent generation is feasible. Additionally, OMT models improve in cross-lingual transfer, being close to solving the “understanding” part of the puzzle in MT for the 1,600 evaluated. Beyond strong out-of-the-box performance, we find that finetuning and retrieval-augmented generation offer additional pathways to improve quality for the given subset of languages when targeted data or domain knowledge is available. Our leaderboard and main humanly created evaluation datasets (BOUQuET and Met-BOUQuET) are dynamically evolving towards Omnilinguality and freely available.</description>
<link>https://ai.meta.com/research/publications/omnilingual-mt-machine-translation-for-1600-languages/</link>
<guid isPermaLink="false">1449924233245450</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual MT Team, Belen Alastruey, Niyati Bafna, Andrea Caciolai, Kevin Heffernan, Artyom Kozhevnikov, Christophe Ropers, Eduardo Sánchez, Charles-Eric Saint-James, Ioannis Tsiamas, Chierh CHENG, Joe Chuang, Paul-Ambroise Duquenne, Mark Duppenthaler, Nate Ekberg, Cynthia Gao, Pere Lluís Huguet Cabot, João Maria Janeiro, Jean Maillard, Gabriel Mejia Gonzalez, Holger Schwenk, Edan Toledo, Arina Turkatenko, Albert Ventayol-Boada, Rashel Moritz, Alexandre Mourachko, Surya Parimi, Mary Williamson, Shireen Yates, David Dale, Marta R. Costa-jussa</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Omnilingual SONAR: Cross-Lingual and Cross-Modal Sentence Embeddings Bridging Massively Multilingual Text and Speech</title>
<description>Cross-lingual sentence encoders have traditionally been limited to a few hundred languages, and have sacrificed downstream performance to achieve better alignment across languages, limiting their adoption. In this work, we introduce OmniSONAR, a novel family of omnilingual, cross-lingual and cross-modal sentence embedding models that breaks this barrier. We establish a unified semantic space, natively encompassing text, speech, code and mathematical expressions, while achieving state-of-the-art downstream performance for an unprecedented scale of thousands of languages, from high-resource languages to extremely low-resource varieties.
To achieve this scale without representation collapse and while maintaining top-tier performance in the high-resource languages, we employ a progressive training strategy. We first build a state-of-the-art foundational embedding space for 200 languages using an LLM-initialized Encoder-Decoder, combining token-level decoding with a novel split-softmax contrastive loss and synthetic hard negatives. Leveraging this strong foundational space, we expand to several thousands of language varieties via a specialized two-stage teacher-student encoder distillation framework. Further modeling extensions derived from OmniSONAR address long context inputs and token-centric representations. Finally, we demonstrate the cross-modal extensibility of this space by seamlessly mapping 177 spoken languages into it.
OmniSONAR redefines the state of the art for multilingual representation learning. It halves the cross-lingual similarity search error rate of the previous best models on the 200 languages of FLORES, while also achieving a staggering 15-fold error rate reduction across 1,560 languages in the BIBLE benchmark. Furthermore, our embedding model enables unprecedented translation capabilities, outperforming NLLB-3B on several multilingual benchmarks, and surpassing all previous models, including multi-billion-parameter LLMs, by 15 chrF++ points in 1,560→English translation in the BIBLE benchmark. Beyond alignment and translation, OmniSONAR demonstrates strong general-purpose capabilities across downstream embedding tasks on MTEB and programming languages on XLCoST. For the speech modality, our massively multilingual extension exhibits a 43% lower error rate in cross-lingual and cross-modal similarity search, while achieving 97% of SeamlessM4T performance in speech-to-text translation, despite being a zero-shot translation model trained only with ASR data. Finally, by training an encoder-decoder language model, Spectrum, exclusively on English text that processes OmniSONAR sequences, we unlock immediate high-performance transfer to thousands of languages and the speech modality for complex downstream tasks. These outstanding results position OmniSONAR as a robust, language- and modality-agnostic foundation for any downstream usage.</description>
<link>https://ai.meta.com/research/publications/omnilingual-sonar-cross-lingual-and-cross-modal-sentence-embeddings-bridging-massively-multilingual-text-and-speech/</link>
<guid isPermaLink="false">1231952259116893</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual SONAR Team, João Maria Janeiro, Pere Lluís Huguet Cabot, Ioannis Tsiamas, Yen Meng, Vivek Iyer, Guillem Ramirez, Loic Barrault, Belen Alastruey, Yu-An Chung, Marta R. Costa-jussa, David Dale, Kevin Heffernan, Jaehyeong Jo, Artyom Kozhevnikov, Alexandre Mourachko, Christophe Ropers, Holger Schwenk, Paul-Ambroise Duquenne</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Speech & Audio</category>
</item>
<item>
<title>Unified Vision–Language Modeling via Concept Space Alignment</title>
<description>We introduce v-Sonar, a vision–language embedding space extended from the text-only embedding space Sonar (Omnilingual Embeddings Team et al., 2026), which supports 1500 text languages and 177 speech languages. To construct v-Sonar, we propose a post-hoc alignment pipeline that maps the representations of an existing vision encoder into the Sonar space. We thoroughly evaluate v-Sonar and show that its embeddings achieve competitive performance on text-to-video retrieval. Equipped with the Sonar text decoder, v-Sonar further surpasses state-of-the-art vision–language models on video captioning tasks, including Dream-1k (Bleu 24.3 vs. 19.6) and Vatex (Bleu 45.0 vs. 41.5).
Leveraging v-Sonar, we first demonstrate that the Large Concept Model (LCM; LCM team et al. 2024) operating in Sonar and trained with English text only, can perform both single- and multi-visual concept understanding in a zero-shot manner. Finally, we introduce v-LCM, which extends the LCM with vision–language instruction tuning. v-LCM encodes vision and language inputs into an unified sequence of latent embeddings via v-Sonar and Sonar, and it is trained with the same latent diffusion objective for next-embedding prediction as in LCM’s text-only pre-training. Experiments on a large-scale multilingual and -modal instruction–tuning data mixture highlight the potential of v-LCM: v-LCM matches state-of-the-art vision-language models on tasks covering image/video captioning and question answering, while significantly outperforming them across 61 rich- to low-resource languages out of all 62 tested languages.</description>
<link>https://ai.meta.com/research/publications/unified-vision-language-modeling-via-concept-space-alignment/</link>
<guid isPermaLink="false">1257213679668978</guid>
<pubDate>Fri, 27 Feb 2026 08:00:00 GMT</pubDate>
<author>Yifu Qiu, Paul-Ambroise Duquenne, Holger Schwenk</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Research</category>
</item>
<item>
<title>Learning Personalized Agents from Human Feedback</title>
<description>Modern AI agents are powerful but often fail to align with the idiosyncratic, evolving preferences of individual users. Prior approaches typically rely on static datasets, either training implicit preference models on interaction history or encoding user profiles in external memory. However, these approaches struggle with new users and with preferences that change over time. We introduce Personalized Agents from Human Feedback (PAHF), a framework for continual personalization in which agents learn online from live interaction using explicit per-user memory. PAHF operationalizes a three-step loop: (1) seeking pre-action clarification to resolve ambiguity, (2) grounding actions in preferences retrieved from memory, and (3) integrating post-action feedback to update memory when preferences drift. To evaluate this capability, we develop a four-phase protocol and two benchmarks in embodied manipulation and online shopping. These benchmarks quantify an agent's ability to learn initial preferences from scratch and subsequently adapt to persona shifts. Our theoretical analysis and empirical results show that integrating explicit memory with dual feedback channels is critical: PAHF learns substantially faster and consistently outperforms both no-memory and single-channel baselines, reducing initial personalization error and enabling rapid adaptation to preference shifts.</description>
<link>https://ai.meta.com/research/publications/learning-personalized-agents-from-human-feedback/</link>
<guid isPermaLink="false">910245944695813</guid>
<pubDate>Thu, 26 Feb 2026 08:00:00 GMT</pubDate>
<author>Kaiqu Liang, Julia Kruk, Shengyi Qian, Xianjun Yang, Shengjie Bi, Shaoliang Nie, Michael Zhang, Lijuan Liu, Jaime Fernández Fisac, Shuyan Zhou, Saghar Hosseini</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Conversational AI</category>
<category>Research</category>
</item>
<item>
<title>FERRET: Framework for Expansion Reliant Red Teaming</title>
<description>We introduce a multi-faceted automated red teaming framework in which the goal is to generate multi-modal adversarial conversations that would break a target model and introduce various expansions that would result in more effective and efficient adversarial conversations. The introduced expansions include: 1. Horizontal expansion in which the goal is for the red team model to self-improve and generate more effective conversation starters that would shape a conversation. 2. Vertical expansion in which the goal is to take these conversation starters that are discovered in the horizontal expansion phase and expand them into effective multi-modal conversations and 3. Meta expansion in which the goal is for the red team model to discover more effective multi-modal attack strategies during the course of a conversation. We call our framework FERRET (Framework for Expansion Reliant Red Teaming) and compare it with various existing automated red teaming approaches. In our experiments, we demonstrate the effectiveness of FERRET in generating effective multi-modal adversarial conversations and its superior performance against existing state of the art approaches.</description>
<link>https://ai.meta.com/research/publications/ferret-framework-for-expansion-reliant-red-teaming/</link>
<guid isPermaLink="false">2396003004185464</guid>
<pubDate>Fri, 13 Feb 2026 08:00:00 GMT</pubDate>
<author>Ninareh Mehrabi, Vítor Albiero, Maya Pavlova, Joanna Bitton</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Responsible AI</category>
</item>
<item>
<title>UniT: Unified Multimodal Chain-of-Thought Test-time Scaling</title>
<description>Unified models can handle both multimodal understanding and generation within a single architecture, yet they typically operate in a single pass without iteratively refining their outputs. Many multimodal tasks, especially those involving complex spatial compositions, multiple interacting objects, or evolving instructions, require decomposing instructions, verifying intermediate results, and making iterative corrections. While test-time scaling (TTS) has demonstrated that allocating additional inference compute for iterative reasoning substantially improves language model performance, extending this paradigm to unified multimodal models remains an open challenge.
We introduce UniT, a framework for multimodal chain-of-thought test-time scaling that enables a single unified model to reason, verify, and refine across multiple rounds. UniT combines agentic data synthesis, unified model training, and flexible test-time inference to elicit cognitive behaviors including verification, subgoal decomposition, and content memory. Our key findings are: (1) unified models trained on short reasoning trajectories generalize to longer inference chains at test time; (2) sequential chain-of-thought reasoning provides a more scalable and compute-efficient TTS strategy than parallel sampling; (3) training on generation and editing trajectories improves out-of-distribution visual reasoning. These results establish multimodal test-time scaling as an effective paradigm for advancing both generation and understanding in unified models.</description>
<link>https://ai.meta.com/research/publications/unit-unified-multimodal-chain-of-thought-test-time-scaling/</link>
<guid isPermaLink="false">1477019817147720</guid>
<pubDate>Wed, 11 Feb 2026 08:00:00 GMT</pubDate>
<author>Leon Liangyu Chen, Haoyu Ma, Zhipeng Fan, Ziqi Huang, Animesh Sinha, Xiaoliang Dai, Jialiang Wang, Zecheng He, Jianwei Yang, Chunyuan Li, Junzhe Sun, Chu Wang, Serena Yeung-Levy, Felix Juefei-Xu</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Computer Vision</category>
</item>
<item>
<title>AIRS-Bench: a Suite of Tasks for Frontier AI Research Science Agents</title>
<description>LLM agents hold significant promise for advancing scientific research. To accelerate this progress, we introduce AIRS-Bench (the AI Research Science Benchmark), a suite of 20 tasks sourced from state-of-the-art machine learning papers. These tasks span diverse domains, including language modeling, mathematics, bioinformatics, and time series forecasting. AIRS-Bench tasks assess agentic capabilities over the full research lifecycle---including idea generation, experiment analysis and iterative refinement---without providing baseline code. The AIRS-Bench task format is versatile, enabling easy integration of new tasks and rigorous comparison across different agentic frameworks. We establish baselines using frontier models paired with both sequential and parallel scaffolds. Our results show that agents exceed human SOTA in four tasks but fail to match it in sixteen others. Even when agents surpass human benchmarks, they do not reach the theoretical performance ceiling for the underlying tasks. These findings indicate that AIRS-Bench is far from saturated and offers substantial room for improvement.
We open-source the AIRS-Bench task definitions and evaluation code to catalyze further development in autonomous scientific research.</description>
<link>https://ai.meta.com/research/publications/airs-bench-a-suite-of-tasks-for-frontier-ai-research-science-agents/</link>
<guid isPermaLink="false">2685677328474642</guid>
<pubDate>Tue, 10 Feb 2026 08:00:00 GMT</pubDate>
<author>Alisia Lupidi, Bhavul Gauri, Thomas Simon Foster, Bassel Al Omari, Despoina Magka, Alberto Pepe, Alexis Audran-Reiss, Muna Aghamelu, Nicolas Baldwin, Lucia Cipolina-Kun, Jean-Christophe Gagnon-Audet, Chee Hau Leow, Sandra Lefdal, Hossam Mossalam, Abhinav Moudgil, Saba Nazir, Emanuel Tewolde, Isabel Urrego, Jordi Armengol-Estape, Amar Budhiraja, Gaurav Chaurasia, Abhishek Charnalia, Derek Dunfield, Karen Hambardzumyan, Daniel Izcovich, Martin Josifoski, Ishita Mediratta, Kelvin Niu, Parth Pathak, Michael Shvartsman, Edan Toledo, Anton Protopopov, Roberta Raileanu, Alexander Miller, Tatiana Shavrina, Jakob Foerster, Yoram Bachrach</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
</item>
<item>
<title>Energy-Based Models for Atomic-Resolution Protein Conformations | Facebook AI Research</title>
<description>We propose an energy-based model (EBM) of protein conformations that operates at atomic scale. The model is trained solely on crystallized protein data.…</description>
<link>https://ai.meta.com/research/publications/energy-based-models-for-atomic-resolution-protein-conformations/</link>
<guid isPermaLink="false">237760337463058</guid>
<pubDate>Sat, 25 Apr 2020 07:00:00 GMT</pubDate>
<author>Yilun Du, Joshua Meier, Jerry Ma, Rob Fergus, Alexander Rives</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
</item>
<item>
<title>Scaling up online speech recognition using ConvNets | Facebook AI Research</title>
<description>We design an online end-to-end speech recognition system based on Time-Depth Separable (TDS) convolutions and Connectionist Temporal Classification (CTC). The system has almost three times the throughput of a well tuned hybrid ASR baseline…</description>
<link>https://ai.meta.com/research/publications/scaling-up-online-speech-recognition-using-convnets/</link>
<guid isPermaLink="false">623727015029499</guid>
<pubDate>Mon, 13 Jan 2020 08:00:00 GMT</pubDate>
<author>Vineel Pratap, Qiantong Xu, Jacob Kahn, Gilad Avidov, Tatiana Likhomanenko, Awni Hannun, Vitaliy Liptchinsky, Gabriel Synnaeve, Ronan Collobert</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Facebook AI's WAT19 Myanmar-English Translation Task Submission</title>
<description>This paper describes Facebook AI's submission to the WAT 2019 Myanmar-English translation task. Our baseline systems are BPE-based transformer models.…</description>
<link>https://ai.meta.com/research/publications/facebook-ai-wat19-myanmar-english-translation-task-submission/</link>
<guid isPermaLink="false">389591045439111</guid>
<pubDate>Thu, 31 Oct 2019 07:00:00 GMT</pubDate>
<author>Peng-Jen Chen, Jiajun Shen, Matt Le, Vishrav Chaudhary, Ahmed El-Kishky, Guillaume Wenzek, Myle Ott, Marc’Aurelio Ranzato</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Order-Aware Generative Modeling Using the 3D-Craft Dataset | Facebook AI Research</title>
<description>Research on 2D and 3D generative models typically focuses on the final artifact being created, e.g., an image or a 3D structure. Unlike 2D image generation, the generation of 3D objects in the real world is commonly constrained by the process…</description>
<link>https://ai.meta.com/research/publications/order-aware-generative-modeling-using-the-3d-craft-dataset/</link>
<guid isPermaLink="false">1247541865430840</guid>
<pubDate>Sun, 27 Oct 2019 07:00:00 GMT</pubDate>
<author>Zhuoyuan Chen, Demi Guo, Tong Xiao, Saining Xie, Xinlei Chen, Haonan Yu, Jonathan Gray, Kavya Srinet, Haoqi Fan, Jerry Ma, Charles R. Qi, Shubham Tulsiani, Arthur Szlam, Larry Zitnick</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
</item>
<item>
<title>ELF OpenGo: An Analysis and Open Reimplementation of AlphaZero | Facebook AI Research</title>
<description>The AlphaGo, AlphaGo Zero, and AlphaZero series of algorithms are remarkable demonstrations of deep reinforcement learning’s capabilities, achieving superhuman performance in the complex game of Go with progressively increasing autonomy.…</description>
<link>https://ai.meta.com/research/publications/elf-opengo-an-analysis-and-open-reimplementation-of-alpha-zero/</link>
<guid isPermaLink="false">2747507891982858</guid>
<pubDate>Tue, 11 Jun 2019 07:00:00 GMT</pubDate>
<author>Yuandong Tian, Jerry Ma, Qucheng Gong, Shubho Sengupta, Zhuoyuan Chen, James Pinkerton, Larry Zitnick</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Computer Vision</category>
</item>
<item>
<title>Equi-normalization of Neural Networks | Facebook AI Research</title>
<description>Modern neural networks are over-parametrized. In particular, each rectified linear hidden unit can be modified by a multiplicative factor by adjusting input and output weights, without changing the rest of the network. Inspired by the…</description>
<link>https://ai.meta.com/research/publications/equi-normalization-of-neural-networks/</link>
<guid isPermaLink="false">2618277071589238</guid>
<pubDate>Sun, 05 May 2019 07:00:00 GMT</pubDate>
<author>Pierre Stock, Benjamin Graham, Rémi Gribonval, Hervé Jégou</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
<category>Computer Vision</category>
</item>
<item>
<title>On the Pitfalls of Measuring Emergent Communication | Facebook AI Research</title>
<description>How do we know if communication is emerging in a multi-agent system? The vast majority of recent papers on emergent communication show that adding a communication channel leads to an increase in reward or task success. This is a useful…</description>
<link>https://ai.meta.com/research/publications/on-the-pitfalls-of-measuring-emergent-communication/</link>
<guid isPermaLink="false">2451938888356370</guid>
<pubDate>Thu, 14 Mar 2019 07:00:00 GMT</pubDate>
<author>Ryan Lowe, Jakob Foerster, Y-Lan Boureau, Joelle Pineau, Yann Dauphin</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Loss in Translation: Learning Bilingual Word Mapping with a Retrieval Criterion | Facebook AI Research</title>
<description>Continuous word representations learned separately on distinct languages can be aligned so that their words become comparable in a common space. Existing works typically solve a least-square regression problem to learn a rotation aligning a…</description>
<link>https://ai.meta.com/research/publications/loss-in-translation-learning-bilingual-word-mapping-with-a-retrieval-criterion/</link>
<guid isPermaLink="false">1432754183558579</guid>
<pubDate>Tue, 30 Oct 2018 07:00:00 GMT</pubDate>
<author>Armand Joulin, Piotr Bojanowski, Tomas Mikolov, Hervé Jégou</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>LAMV: Learning to align and match videos with kernelized temporal layers | Facebook AI Research</title>
<description>This paper considers a learnable approach for comparing and aligning videos. Our architecture builds upon and revisits temporal match kernels within neural networks: we propose a new temporal layer that finds temporal alignments by maximizing…</description>
<link>https://ai.meta.com/research/publications/lamv-learning-to-align-and-match-videos-with-kernelized-temporal-layers/</link>
<guid isPermaLink="false">2177100492586051</guid>
<pubDate>Tue, 19 Jun 2018 07:00:00 GMT</pubDate>
<author>Lorenzo Baraldi, Matthijs Douze, Rita Cucchiara, Hervé Jégou</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Computer Vision</category>
<category>Research</category>
</item>
<item>
<title>Low-shot learning with large-scale diffusion | Facebook AI Research</title>
<description>This paper considers the problem of inferring image labels from images when only a few annotated examples are available at training time. This setup is often referred to as low-shot learning, where a standard approach is to re-train the last…</description>
<link>https://ai.meta.com/research/publications/low-shot-learning-with-large-scale-diffusion/</link>
<guid isPermaLink="false">534360120449321</guid>
<pubDate>Mon, 18 Jun 2018 07:00:00 GMT</pubDate>
<author>Matthijs Douze, Arthur Szlam, Bharath Hariharan, Hervé Jégou</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Speech & Audio</category>
<category>Computer Vision</category>
<category>Research</category>
</item>
<item>
<title>Mastering the Dungeon: Grounded Language Learning by Mechanical Turker Descent | Facebook AI Research</title>
<description>Contrary to most natural language processing research, which makes use of static datasets, humans learn language interactively, grounded in an environment. In this work we propose an interactive learning procedure called Mechanical…</description>
<link>https://ai.meta.com/research/publications/mastering-the-dungeon-grounded-language-learning-by-mechanical-turker-descent/</link>
<guid isPermaLink="false">2704786416424529</guid>
<pubDate>Mon, 30 Apr 2018 07:00:00 GMT</pubDate>
<author>Zhilin Yang, Saizheng Zhang, Jack Urbanek, Will Feng, Alexander H. Miller, Arthur Szlam, Douwe Kiela, Jason Weston</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
<category>Computer Vision</category>
</item>
<item>
<title>Polysemous Codes | Facebook AI Research</title>
<description>This paper considers the problem of approximate nearest neighbor search in the compressed domain. We introduce polysemous codes, which offer both the distance estimation quality of product quantization and the efficient comparison of binary…</description>
<link>https://ai.meta.com/research/publications/polysemous-codes/</link>
<guid isPermaLink="false">617678119003343</guid>
<pubDate>Mon, 10 Oct 2016 07:00:00 GMT</pubDate>
<author>Matthijs Douze, Hervé Jégou, Florent Perronnin</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=Yu1CX1Zow78nWBKqKpNd7w&_nc_ss=73289&oh=00_Af3Vqoyo4Cd2n-7DLihU_yg21y0yvk0CoMMa9fXn_9luaw&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Speech & Audio</category>
<category>Computer Vision</category>
<category>Research</category>
</item>
</channel>
</rss>http://localhost:1200/meta/ai/global-search/content_types=dataset - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — content_types=dataset</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/content_types=dataset" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/. - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:22:39 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>Multi-room Apartments Simulation (MRAS) Dataset | Meta AI Research</title>
<description>The Multi-Room Apartments Simulation (MRAS) dataset is a multi-modal dataset created for the task of estimating spatially-distributed acoustic parameters in complex scenes.</description>
<link>https://ai.meta.com/datasets/mras-simulated-rooms-dataset/</link>
<guid isPermaLink="false">2247220155683256</guid>
<pubDate>Wed, 28 Jan 2026 18:49:18 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/612410014_2220471548441535_606775622330495537_n.jpg?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=1bE1l45d6ZQQ7kNvwGzExIi&_nc_oc=Ado-6Ct9Xx3eWjk90FaKNEtJode5O_FsHbdvIUqlBg0pmSXdVV101WYNtSFvKg9wQLc&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=ak5zaINxnmY9YYI-FX5x3Q&_nc_ss=73289&oh=00_Af0wN3vaPR5HDEXjSNkBAaNVdEYb27HVIVtbVR--AiuhJQ&oe=6A08CB19" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
<item>
<title>Meta Synthetic Environments Lidar Dataset | Meta AI Research</title>
<description>The Meta Synthetic Environments (MSE) Lidar Dataset is the first-of-its-kind large-scale single-photon lidar dataset, built on top of Aria Synthetic Environments (ASE) and intended to unlock new machine learning capabilities for single-photon lidars.</description>
<link>https://ai.meta.com/datasets/meta-synthetic-environments-lidar-dataset/</link>
<guid isPermaLink="false">1350377946838590</guid>
<pubDate>Wed, 17 Dec 2025 17:38:04 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/579919244_838469262096799_1005035704573286504_n.jpg?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=o9JbB5ingBsQ7kNvwHzZ9iJ&_nc_oc=AdpkC6WaZ5xyfWpfOtMXdU3SlyaRRd3d6qq-XPihTulVL4E_g4xaXXvv8h7Isjk-udY&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=ak5zaINxnmY9YYI-FX5x3Q&_nc_ss=73289&oh=00_Af3gReQI9hFksZMGNGlxNoM3g8iMOByYx0vXxnPfeM8FBA&oe=6A08C2C4" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
<item>
<title>PLM Data | Meta AI Research</title>
<description>PLM Data is a comprehensive collection of synthetic and human-annotated datasets for detailed visual understanding, combining existing and newly collected data for tasks like OCR, captioning, and visual question answering.</description>
<link>https://ai.meta.com/datasets/plm-data/</link>
<guid isPermaLink="false">672877175127332</guid>
<pubDate>Tue, 27 May 2025 16:18:53 GMT</pubDate>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.2365-6/384127954_1010009116964035_2906046058102555347_n.jpg?_nc_cat=101&ccb=1-7&_nc_sid=e280be&_nc_ohc=iT1pu9-tiuUQ7kNvwEbfCoY&_nc_oc=Adr-u-AU0EGWmbHpZtKbA6KmkwzhWh2AFteenLAyiggrPjqxDhdcJ5FaCgcQkaFNHkM&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=ak5zaINxnmY9YYI-FX5x3Q&_nc_ss=73289&oh=00_Af3766_b68AyzSfMuw_DaXM4MJFqXY-gfAzJY-K8wp_i4w&oe=6A08DD05" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
<item>
<title>PE Video | Meta AI Research</title>
<description>The PE Video Dataset (PVD) is a large-scale collection of 1 million diverse videos, featuring 120,000+ expertly annotated clips. The dataset was introduced in our paper "Perception Encoder".</description>
<link>https://ai.meta.com/datasets/pe-video/</link>
<guid isPermaLink="false">29433953799551622</guid>
<pubDate>Thu, 17 Apr 2025 18:01:37 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/491815770_995150272810861_1471459038704351685_n.jpg?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=Yffaq2cg7KYQ7kNvwEYAtHj&_nc_oc=Adph248-a3vlrkmqPBRH-Jg5Os5r--KB69-Ekumlkn2xPqzcHDR_AMujThkKe6h0Ugg&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=ak5zaINxnmY9YYI-FX5x3Q&_nc_ss=73289&oh=00_Af0M3KEOTwiknMprd5ZaQp3BDujpPT76TZV9qq3qRizI6Q&oe=6A08D0C8" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
</channel>
</rss>... |
Contributor
http://localhost:1200/meta/ai/global-search/content_types=tool - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — content_types=tool</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/content_types=tool" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/. - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:22:43 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>FastText</title>
<description>A lightweight library designed to help build scalable solutions for text representation and classification.</description>
<link>https://ai.meta.com/tools/fasttext/</link>
<guid isPermaLink="false">842058939461581</guid>
<pubDate>Tue, 14 Oct 2025 23:29:35 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=bDMhbM6-lILsBveZ8Vuxvg&_nc_ss=73289&oh=00_Af0rP__-9Q7gX_RD7dYIa3C5aPvzYEu-UjYyoxqxRrWa4A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>DensePose</title>
<description>DensePose, dense human pose estimation, is designed to map all human pixels of an RGB image to a 3D surface-based representation of the human body.</description>
<link>https://ai.meta.com/tools/densepose/</link>
<guid isPermaLink="false">368589657320353</guid>
<pubDate>Tue, 14 Oct 2025 22:00:22 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=bDMhbM6-lILsBveZ8Vuxvg&_nc_ss=73289&oh=00_Af0rP__-9Q7gX_RD7dYIa3C5aPvzYEu-UjYyoxqxRrWa4A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>KILT Benchmarking</title>
<description>KILT is a resource for training, evaluating and analyzing NLP models on Knowledge Intensive Language Tasks.</description>
<link>https://ai.meta.com/tools/kilt/</link>
<guid isPermaLink="false">2743325952604548</guid>
<pubDate>Thu, 11 Sep 2025 21:42:53 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/119073509_368822947476971_6922122926535022194_n.png?_nc_cat=106&ccb=1-7&_nc_sid=e280be&_nc_ohc=Mg93n9KMtaEQ7kNvwEKlDOD&_nc_oc=Adop-U6wmtrO1PSZN9qX6NsI3Ktzl9lY8L5KpmF8KTBYtwMnY4wlFu3C82RqFxapcRQ&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=bDMhbM6-lILsBveZ8Vuxvg&_nc_ss=73289&oh=00_Af1u_WaiT8dTnEksXH1GN3sbXaTmBmJTNQoMeQdzwAjfdg&oe=6A08CF54" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>Faiss</title>
<description>A library that allows developers to quickly search for embeddings of multimedia documents that are similar to each other.</description>
<link>https://ai.meta.com/tools/faiss/</link>
<guid isPermaLink="false">297198177617190</guid>
<pubDate>Wed, 05 Feb 2025 00:52:31 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=bDMhbM6-lILsBveZ8Vuxvg&_nc_ss=73289&oh=00_Af0rP__-9Q7gX_RD7dYIa3C5aPvzYEu-UjYyoxqxRrWa4A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>System Cards - Meta AI</title>
<description>System cards or multiple machine learning models help users understand the intention, impact and limitations of our AI systems.</description>
<link>https://ai.meta.com/tools/system-cards/</link>
<guid isPermaLink="false">632380378064785</guid>
<pubDate>Tue, 12 Dec 2023 18:58:53 GMT</pubDate>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.2365-6/343413316_1234589910516703_2524736655913484825_n.jpg?_nc_cat=101&ccb=1-7&_nc_sid=e280be&_nc_ohc=IaCyxndRw5IQ7kNvwEPF5kL&_nc_oc=Ados5yqqUSc56_IvQLx3_CeH5K-fY0LWwr4R-Famo0MmyH_w68wqkY1TopL3n4wLQ6U&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=bDMhbM6-lILsBveZ8Vuxvg&_nc_ss=73289&oh=00_Af1nEPt7D2LCVuBtxFnZxEF84nwqV8a_ClYf5Jk7kCZ8Qw&oe=6A08D680" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>Segment Anything Dataset</title>
<description>Segment Anything 1 Billion (SA-1B) is a dataset designed for training general-purpose object segmentation models from open world images. The dataset was introduced in our paper “Segment Anything”.</description>
<link>https://ai.meta.com/tools/segment-anything-dataset/</link>
<guid isPermaLink="false">1375631516311803</guid>
<pubDate>Tue, 25 Jul 2023 01:17:09 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=bDMhbM6-lILsBveZ8Vuxvg&_nc_ss=73289&oh=00_Af0rP__-9Q7gX_RD7dYIa3C5aPvzYEu-UjYyoxqxRrWa4A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>PyTorch</title>
<description>PyTorch is an open source deep learning framework built to be flexible and modular for research, with the stability and support needed for production deployment.</description>
<link>https://ai.meta.com/tools/pytorch/</link>
<guid isPermaLink="false">297718674239828</guid>
<pubDate>Wed, 08 Mar 2023 02:37:08 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=bDMhbM6-lILsBveZ8Vuxvg&_nc_ss=73289&oh=00_Af0rP__-9Q7gX_RD7dYIa3C5aPvzYEu-UjYyoxqxRrWa4A&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
<item>
<title>Casual Conversations V2 Dataset</title>
<description>Casual Conversations V2 dataset is designed to measure the robustness of AI models across a diverse set of age, genders, apparent skin tones and ambient lighting conditions.</description>
<link>https://ai.meta.com/tools/casual-conversations-v2-dataset/</link>
<guid isPermaLink="false">1646023009169056</guid>
<pubDate>Tue, 07 Mar 2023 23:02:56 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/335035828_927765094902963_5055688398222102698_n.jpg?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=7pNQPMjhOWgQ7kNvwGMCW-W&_nc_oc=AdpAK1kmc5KSdzJxnSlPYv3i7gBO035Rh79lfBroyK2mXyW69Gxac1ruEdRfo0ymnpU&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=bDMhbM6-lILsBveZ8Vuxvg&_nc_ss=73289&oh=00_Af2naaGBm8dTmpOBRi71GU1qQdafEXjb7QeuY85cq8MydA&oe=6A08B7BC" type="image/jpeg"></enclosure>
<category>TOOL</category>
</item>
</channel>
</rss>... |
Contributor
http://localhost:1200/meta/ai/global-search/sort_by=MOST_RECENT - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/sort_by=MOST_RECENT" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/. - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:22:49 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>AIRA₂: Overcoming Bottlenecks in AI Research Agents</title>
<description>Existing research has identified three structural performance bottlenecks in AI research agents: (1) synchronous single-GPU execution constrains sample throughput, limiting the benefit of search; (2) a generalization gap where validation-based selection causes overfitting and performance to degrade over extended search horizons; and (3) the limited capability of fixed, single-turn LLM operators imposes a ceiling on search performance. We introduce AIRA₂, which addresses these bottlenecks through three architectural choices: an asynchronous multi-GPU worker pool that increases experiment throughput linearly; a Hidden Consistent Evaluation protocol that delivers a reliable evaluation signal; and ReAct agents that dynamically scope their actions and debug interactively. On MLE-bench-30, AIRA₂ achieves a mean Percentile Rank of 81.5% at 24 hours and 83.1% at 72 hours, outperforming the strongest baseline, which achieves 72.7%. On AIRS-Bench, AIRA₂ exceeds human state-of-the-art on 6 out of 20 diverse research tasks. Ablations confirm that each architectural component is necessary, that performance follows a predictable scaling law that transfers across LLM backbones, and that the "overfitting" reported in prior work was driven by evaluation noise rather than true data memorization.</description>
<link>https://ai.meta.com/research/publications/aira-overcoming-bottlenecks-in-ai-research-agents/</link>
<guid isPermaLink="false">1298625375513718</guid>
<pubDate>Thu, 16 Apr 2026 07:00:00 GMT</pubDate>
<author>Karen Hambardzumyan, Nicolas Baldwin, Edan Toledo, Rishi Hazra, Michael Kuchnik, Bassel Al Omari, Thomas Simon Foster, Anton Protopopov, Jean-Christophe Gagnon-Audet, Ishita Mediratta, Kelvin Niu, Michael Shvartsman, Alisia Lupidi, Alexis Audran-Reiss, Parth Pathak, Tatiana Shavrina, Despoina Magka, Hela Momand, Derek Dunfield, Nicola Cancedda, Pontus Stenetorp, Carole-Jean Wu, Jakob Foerster, Yoram Bachrach, Martin Josifoski</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
</item>
<item>
<title>TransText: Transparency Aware Image-to-Video Typography Animation</title>
<description>We introduce the first method, to the best of our knowledge, for adapting image-to-video models to layer-aware text (glyph) animation, a capability critical for practical dynamic visual design. Existing approaches predominantly handle the transparency-encoding (alpha channel) as an extra latent dimension appended to the RGB space, necessitating the reconstruction of the underlying RGB-centric variational autoencoder (VAE). However, given the scarcity of high-quality transparent glyph data, retraining the VAE is computationally expensive and may erode the robust semantic priors learned from massive RGB corpora, potentially leading to latent pattern mixing. To mitigate these limitations, we propose TransText, a framework based on a novel Alpha-as-RGB paradigm to jointly model appearance and transparency without modifying the pre-trained generative manifold. TransText embeds the alpha channel as an RGB-compatible visual signal through latent spatial concatenation, explicitly ensuring strict cross-modal (RGB-and-Alpha) consistency while preventing feature entanglement. Our experiments demonstrate that TransText significantly outperforms baselines, generating coherent, high-fidelity transparent animations with diverse, fine-grained effects.</description>
<link>https://ai.meta.com/research/publications/transtext-transparency-aware-image-to-video-typography-animation/</link>
<guid isPermaLink="false">2172593836888174</guid>
<pubDate>Tue, 14 Apr 2026 07:00:00 GMT</pubDate>
<author>Fei Zhang, Zijian Zhou, Bohao Tang, Sen He, Hang Li (BizAI), Zhe Wang, Soubhik Sanyal, Pengfei Liu, Viktar Atliha, Tao Xiang, Frost Xu, Semih Gunel</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Computer Vision</category>
<category>ML Applications</category>
</item>
<item>
<title>Think in Strokes, Not Pixels: Process-Driven Image Generation via Interleaved Reasoning</title>
<description>Humans paint images incrementally: they plan a global layout, sketch a coarse draft, inspect, and refine details, and most importantly, each step is grounded in the evolving visual states. However, can unified multimodal models trained on text-image interleaved datasets also imagine the chain of intermediate states? In this paper, we introduce process-driven image generation, a multi-step paradigm that decomposes synthesis into an interleaved reasoning trajectory of thoughts and actions. Rather than generating images in a single step, our approach unfolds across multiple iterations, each consisting of 4 stages: textual planning, visual drafting, textual reflection, and visual refinement. The textual reasoning explicitly conditions how the visual state should evolve, while the generated visual intermediate in turn constrains and grounds the next round of textual reasoning. A core challenge of process-driven generation stems from the ambiguity of intermediate states: how can models evaluate each partially-complete image? We address this through dense, step-wise supervision that maintains two complementary constraints: for the visual intermediate states, we enforce the spatial and semantic consistency; for the textual intermediate states, we preserve the prior visual knowledge while enabling the model to identify and correct prompt-violating elements. This makes the generation process explicit, interpretable, and directly supervisable. To validate proposed method, we conduct experiments under various text-to-image generation benchmarks.</description>
<link>https://ai.meta.com/research/publications/think-in-strokes-not-pixels-process-driven-image-generation-via-interleaved-reasoning/</link>
<guid isPermaLink="false">847030005080644</guid>
<pubDate>Thu, 09 Apr 2026 07:00:00 GMT</pubDate>
<author>Lei Zhang, Junjiao Tian, Zhipeng Fan, Kunpeng Li, Jialiang Wang, Weifeng Chen, Markos Georgopoulos, Felix Xu, Yuxiao Bao, Julian McAuley, Manling Li, Zecheng He</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Computer Vision</category>
</item>
<item>
<title>Scaling How We Build and Test Our Most Advanced AI</title>
<description>As we build more capable, personalized AI, reliability, security, and user protections are more important than ever. </description>
<link>https://ai.meta.com/blog/scaling-how-we-build-test-advanced-ai/</link>
<guid isPermaLink="false">4369469056623958</guid>
<pubDate>Wed, 08 Apr 2026 17:00:17 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/668130848_1453339552354697_3291493920555502062_n.png?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=yvFttOBoRYkQ7kNvwHAGoUI&_nc_oc=AdrMTKarV4pYb-C6Lo7cZ4FS8VtfHNic415DSbyvO9pcVHrt8wGk3pW7eGF_3ATLqB8&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af02edxBkQI3tqqUZpyn7lXp0xS5no1wJ00GD25nYo2G9g&oe=6A08CDEE" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Introducing Muse Spark: Scaling Towards Personal Superintelligence</title>
<description></description>
<link>https://ai.meta.com/blog/introducing-muse-spark-msl/</link>
<guid isPermaLink="false">963191949487556</guid>
<pubDate>Wed, 08 Apr 2026 16:00:59 GMT</pubDate>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.2365-6/667507363_1871749950191919_4996140970371485757_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=IbntS1cBAcsQ7kNvwEkZk-q&_nc_oc=Ado4EZgw05h6efdlgHKUmEWqiQAUivkkK5wab1s8qYcxECiinRzqsBcSp7BtuIVOo5k&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af3bvvv-nzIowVHaka7NHzpQ5IddD22576-2X_KuRC52DA&oe=6A08D22C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>How Alta Daily Uses Meta’s Segment Anything to Reimagine the Digital Closet</title>
<description></description>
<link>https://ai.meta.com/blog/alta-daily-fashion-app-segment-anything/</link>
<guid isPermaLink="false">2194788291268065</guid>
<pubDate>Mon, 06 Apr 2026 18:53:27 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/658967078_993415586351079_325919381939729048_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=cBKF2Wk0XYoQ7kNvwGT68WW&_nc_oc=Adr_OUg82eUw0EbluNihH00lRvwhw0cI82roesUhcZVs9_YWPOnE24LdOGDODhZ8PXE&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1VT6D43FxT78z1TI8hB4NJz1Uuf4Wq-jaDX6B8snDA9g&oe=6A08B252" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning</title>
<description></description>
<link>https://ai.meta.com/blog/segment-anything-model-3/</link>
<guid isPermaLink="false">715047027731045</guid>
<pubDate>Fri, 27 Mar 2026 16:00:37 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/583339165_852765810541557_5592130597432767690_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=WvtkkpGzh44Q7kNvwHtq_BU&_nc_oc=AdqOSUbFE1HjLhxvY58CCZUsdmat-y6v9A-cIg_tkySMD-A2NT_eE5pWvyzi6Gtrlq0&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1gbycd8pjKJU-sQaqtaJR8JFTVL6k7D-Ag0C7fSCMB9A&oe=6A08D061" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>Introducing TRIBE v2: A Predictive Foundation Model Trained to Understand How the Human Brain Processes Complex Stimuli</title>
<description></description>
<link>https://ai.meta.com/blog/tribe-v2-brain-predictive-foundation-model/</link>
<guid isPermaLink="false">25512119895129580</guid>
<pubDate>Thu, 26 Mar 2026 16:00:16 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/659112495_952140057346056_1728296311373871062_n.gif?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=rBe_VZxcFD4Q7kNvwEl0TGK&_nc_oc=Adq6bzVjbryLAhis8bGJPU1Umwebg8J3Ul_Sn-xoSPOJfZTVx6fNEnrsqhxPP2J05wc&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1UtP53535o1wNMpX_rngyKKOjAylZlSW2noNdOqCbrdg&oe=6A08AF6B" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Research</category>
</item>
<item>
<title>A foundation model of vision, audition, and language for in-silico neuroscience</title>
<description>Cognitive neuroscience is fragmented into specialized models, each tailored to specific experimental paradigms, hence preventing a unified model of cognition in the human brain. Here, we introduce TRIBE v2, a tri-modal (video, audio and language) foundation model capable of predicting human brain activity in a variety of naturalistic and experimental conditions. Leveraging a unified dataset of over 1,000 hours of fMRI across 720 subjects, we demonstrate that our model accurately predicts high-resolution brain responses for novel stimuli, tasks and subjects, superseding traditional linear encoding models, delivering several-fold improvements in accuracy. Critically, TRIBE v2 enables in silico experimentation: tested on seminal visual and neuro-linguistic paradigms, it recovers a variety of results established by decades of empirical research. Finally, by extracting interpretable latent features, TRIBE v2 reveals the fine-grained topography of multisensory integration. These results establish artificial intelligence as a unifying framework for exploring the functional organization of the human brain.</description>
<link>https://ai.meta.com/research/publications/a-foundation-model-of-vision-audition-and-language-for-in-silico-neuroscience/</link>
<guid isPermaLink="false">1009839085553251</guid>
<pubDate>Thu, 26 Mar 2026 07:00:00 GMT</pubDate>
<author>Stéphane d'Ascoli, Jérémy Rapin, Yohann Benchetrit, Teon Brooks, Katelyn Begany, Josephine Raugel, Hubert Jacob Banville, Jean Remi King</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
</item>
<item>
<title>HyperAgents</title>
<description>Self-improving AI systems aim to reduce reliance on human engineering by learning to improve their own learning and problem-solving processes. Existing approaches to recursive self-improvement typically rely on fixed, handcrafted meta-level mechanisms, which fundamentally limit how fast such systems can improve. The Darwin Gödel Machine (DGM)(Zhang et al., 2025b) demonstrates that open-ended self-improvement is achievable in coding. Starting from a single coding agent, the DGM repeatedly generates and evaluates self-modified variants, forming a growing archive of stepping stones for future improvement. Because both evaluation and self-modification are coding tasks, gains in coding ability can translate into gains in self-improvement ability. However, this alignment does not generally hold beyond coding domains.
We introduce hyperagents, self-referential agents that integrate a task agent (which solves the target task) and a meta agent (which modifies itself and the task agent) into a single editable program. Crucially, the meta-level modification procedure is itself editable, enabling metacognitive self-modification, improving not only task-solving behavior, but also the mechanism that generates future improvements. We instantiate this framework by extending DGM to create DGM-Hyperagents (DGM-H). By allowing the improvement procedure to evolve, the DGM-H eliminates the assumption of domain-specific alignment between task performance and self-modification skill, and can potentially support self-accelerating progress on any computable task. Across diverse domains (coding, paper review, robotics reward design, and Olympiad-level math-solution grading), the DGM-H improves performance over time and outperforms baselines without self-improvement or open-ended exploration, as well as prior self-improving systems like DGM. We further show that the DGM-H improves the process by which it generates new agents (e.g., persistent memory, performance tracking), and that these meta-level improvements transfer across domains and accumulate across runs. All experiments were conducted with safety precautions (e.g., sandboxing, human oversight). We discuss what safety entails in this setting and the broader implications of self-improving systems. DGM-Hyperagents offer a glimpse of open-ended AI systems that do not merely search for better solutions, but continually improve their search for how to improve.</description>
<link>https://ai.meta.com/research/publications/hyperagents/</link>
<guid isPermaLink="false">1669124537439877</guid>
<pubDate>Tue, 24 Mar 2026 07:00:00 GMT</pubDate>
<author>Jenny Zhang, Bingchen Zhao, Winnie Yang, Jakob Foerster, Sam Devlin, Tatiana Shavrina</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
<category>Open Source</category>
</item>
<item>
<title>Omnilingual MT: Machine Translation for 1,600 Languages</title>
<description>Advances made through No Language Left Behind (NLLB) have demonstrated that high-quality machine translation (MT) scale to 200 languages. Later Large Language Models (LLMs) have been adopted for MT, increasing in quality but not necessarily extending language coverage. Current systems remain constrained by limited coverage and a persistent generation bottleneck: while crosslingual transfer enables models to somehow understand many undersupported languages, they often cannot generate them reliably, leaving most of the world’s 7,000 languages—especially endangered and marginalized ones—outside the reach of modern MT. Early explorations in extreme scaling offered promising proofs of concept but did not yield sustained solutions.
We present Omnilingual Machine Translation (OMT), the first MT system supporting more than 1,600 languages. This scale is enabled by a comprehensive data strategy that integrates large public multilingual corpora with newly created datasets, including manually curated MeDLEY bitext, synthetic backtranslation, and mining, substantially expanding coverage across long-tail languages, domains, and registers. To ensure both reliable and expansive evaluation, we combined standard metrics with a suite of evaluation artifacts: BLASER 3 quality estimation model (reference-free), OmniTOX toxicity classifier, BOUQuET dataset (a newly created, largest-to-date multilingual evaluation collection built from scratch and manually extended across a wide range of linguistic families), and Met-BOUQuET dataset (faithful multilingual quality estimation at scale). We explore two ways of specializing an LLM for machine translation: as a decoder-only model (OMT-LLaMA) or as a module in an encoder–decoder architecture (OMT-NLLB). The former consists of a model built on LLaMA3, with multilingual continual pretraining and retrieval-augmented translation for inference-time adaptation. The latter is a model built on top of a multilingual aligned space (OmniSONAR, itself also based on LLaMA3), and introduces a training methodology that can exploit non-parallel data, allowing us to incorporate the decoder-only continuous pretraining data into the training of an encoder–decoder architecture. Notably, all our 1B to 8B parameter models match or exceed the MT performance of a 70B LLM baseline, revealing a clear specialization advantage and enabling strong translation quality in low-compute settings. Moreover, our evaluation of English-to-1,600 translations further shows that while baseline models can interpret undersupported languages, they frequently fail to generate them with meaningful fidelity; OMT-LLaMA models substantially expand the set of languages for which coherent generation is feasible. Additionally, OMT models improve in cross-lingual transfer, being close to solving the “understanding” part of the puzzle in MT for the 1,600 evaluated. Beyond strong out-of-the-box performance, we find that finetuning and retrieval-augmented generation offer additional pathways to improve quality for the given subset of languages when targeted data or domain knowledge is available. Our leaderboard and main humanly created evaluation datasets (BOUQuET and Met-BOUQuET) are dynamically evolving towards Omnilinguality and freely available.</description>
<link>https://ai.meta.com/research/publications/omnilingual-mt-machine-translation-for-1600-languages/</link>
<guid isPermaLink="false">1449924233245450</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual MT Team, Belen Alastruey, Niyati Bafna, Andrea Caciolai, Kevin Heffernan, Artyom Kozhevnikov, Christophe Ropers, Eduardo Sánchez, Charles-Eric Saint-James, Ioannis Tsiamas, Chierh CHENG, Joe Chuang, Paul-Ambroise Duquenne, Mark Duppenthaler, Nate Ekberg, Cynthia Gao, Pere Lluís Huguet Cabot, João Maria Janeiro, Jean Maillard, Gabriel Mejia Gonzalez, Holger Schwenk, Edan Toledo, Arina Turkatenko, Albert Ventayol-Boada, Rashel Moritz, Alexandre Mourachko, Surya Parimi, Mary Williamson, Shireen Yates, David Dale, Marta R. Costa-jussa</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>NLP</category>
</item>
<item>
<title>Omnilingual SONAR: Cross-Lingual and Cross-Modal Sentence Embeddings Bridging Massively Multilingual Text and Speech</title>
<description>Cross-lingual sentence encoders have traditionally been limited to a few hundred languages, and have sacrificed downstream performance to achieve better alignment across languages, limiting their adoption. In this work, we introduce OmniSONAR, a novel family of omnilingual, cross-lingual and cross-modal sentence embedding models that breaks this barrier. We establish a unified semantic space, natively encompassing text, speech, code and mathematical expressions, while achieving state-of-the-art downstream performance for an unprecedented scale of thousands of languages, from high-resource languages to extremely low-resource varieties.
To achieve this scale without representation collapse and while maintaining top-tier performance in the high-resource languages, we employ a progressive training strategy. We first build a state-of-the-art foundational embedding space for 200 languages using an LLM-initialized Encoder-Decoder, combining token-level decoding with a novel split-softmax contrastive loss and synthetic hard negatives. Leveraging this strong foundational space, we expand to several thousands of language varieties via a specialized two-stage teacher-student encoder distillation framework. Further modeling extensions derived from OmniSONAR address long context inputs and token-centric representations. Finally, we demonstrate the cross-modal extensibility of this space by seamlessly mapping 177 spoken languages into it.
OmniSONAR redefines the state of the art for multilingual representation learning. It halves the cross-lingual similarity search error rate of the previous best models on the 200 languages of FLORES, while also achieving a staggering 15-fold error rate reduction across 1,560 languages in the BIBLE benchmark. Furthermore, our embedding model enables unprecedented translation capabilities, outperforming NLLB-3B on several multilingual benchmarks, and surpassing all previous models, including multi-billion-parameter LLMs, by 15 chrF++ points in 1,560→English translation in the BIBLE benchmark. Beyond alignment and translation, OmniSONAR demonstrates strong general-purpose capabilities across downstream embedding tasks on MTEB and programming languages on XLCoST. For the speech modality, our massively multilingual extension exhibits a 43% lower error rate in cross-lingual and cross-modal similarity search, while achieving 97% of SeamlessM4T performance in speech-to-text translation, despite being a zero-shot translation model trained only with ASR data. Finally, by training an encoder-decoder language model, Spectrum, exclusively on English text that processes OmniSONAR sequences, we unlock immediate high-performance transfer to thousands of languages and the speech modality for complex downstream tasks. These outstanding results position OmniSONAR as a robust, language- and modality-agnostic foundation for any downstream usage.</description>
<link>https://ai.meta.com/research/publications/omnilingual-sonar-cross-lingual-and-cross-modal-sentence-embeddings-bridging-massively-multilingual-text-and-speech/</link>
<guid isPermaLink="false">1231952259116893</guid>
<pubDate>Tue, 17 Mar 2026 07:00:00 GMT</pubDate>
<author>Omnilingual SONAR Team, João Maria Janeiro, Pere Lluís Huguet Cabot, Ioannis Tsiamas, Yen Meng, Vivek Iyer, Guillem Ramirez, Loic Barrault, Belen Alastruey, Yu-An Chung, Marta R. Costa-jussa, David Dale, Kevin Heffernan, Jaehyeong Jo, Artyom Kozhevnikov, Alexandre Mourachko, Christophe Ropers, Holger Schwenk, Paul-Ambroise Duquenne</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Speech & Audio</category>
</item>
<item>
<title>Four MTIA Chips in Two Years: Scaling AI Experiences for Billions</title>
<description>Serving a wide range of AI models on a global scale, while maintaining the lowest possible costs, is one of the most demanding infrastructure challenges in the industry. </description>
<link>https://ai.meta.com/blog/meta-mtia-scale-ai-chips-for-billions/</link>
<guid isPermaLink="false">814227801000565</guid>
<pubDate>Wed, 11 Mar 2026 14:00:55 GMT</pubDate>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/649719916_2147323009439466_6866394022243367245_n.png?_nc_cat=103&ccb=1-7&_nc_sid=e280be&_nc_ohc=HQUtIT-12s8Q7kNvwGOLZwg&_nc_oc=AdrxxwjcxH_agF9UMFF-AQWK2vy12eCF5p5jf1ozY-qI61a_wgoP7vvJizTyzRxV0xc&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af2Ka9sJU4TKRom4oMYCzkMYS6FRMMiCF2hWY8WsG9ocYQ&oe=6A08CB6C" type="image/jpeg"></enclosure>
<category>BLOG</category>
</item>
<item>
<title>Mapping the World's Forests with Greater Precision: Introducing Canopy Height Maps v2</title>
<description>In partnership with the World Resources Institute, today we’re announcing Canopy Height Maps v2 (CHMv2), an open source model, along with world-scale maps generated with the model.</description>
<link>https://ai.meta.com/blog/world-resources-institute-dino-canopy-height-maps-v2/</link>
<guid isPermaLink="false">1675459620286410</guid>
<pubDate>Tue, 10 Mar 2026 16:00:06 GMT</pubDate>
<enclosure url="https://scontent-hkg4-2.xx.fbcdn.net/v/t39.2365-6/644822799_1500169394861918_3213642176014981812_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=5tvZOejkl8MQ7kNvwE3rzOr&_nc_oc=AdpUWj25o490QP_XxwEWiK2CqMWOSxEQWHDJ-K-g7z3HzWbeYKo0AeH72yUz_8SC2Pk&_nc_zt=14&_nc_ht=scontent-hkg4-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af136IlJfC6fon92xIZ8geSsV0quecM4_DJEQBuhGtOgqw&oe=6A08D606" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
<item>
<title>Unified Vision–Language Modeling via Concept Space Alignment</title>
<description>We introduce v-Sonar, a vision–language embedding space extended from the text-only embedding space Sonar (Omnilingual Embeddings Team et al., 2026), which supports 1500 text languages and 177 speech languages. To construct v-Sonar, we propose a post-hoc alignment pipeline that maps the representations of an existing vision encoder into the Sonar space. We thoroughly evaluate v-Sonar and show that its embeddings achieve competitive performance on text-to-video retrieval. Equipped with the Sonar text decoder, v-Sonar further surpasses state-of-the-art vision–language models on video captioning tasks, including Dream-1k (Bleu 24.3 vs. 19.6) and Vatex (Bleu 45.0 vs. 41.5).
Leveraging v-Sonar, we first demonstrate that the Large Concept Model (LCM; LCM team et al. 2024) operating in Sonar and trained with English text only, can perform both single- and multi-visual concept understanding in a zero-shot manner. Finally, we introduce v-LCM, which extends the LCM with vision–language instruction tuning. v-LCM encodes vision and language inputs into an unified sequence of latent embeddings via v-Sonar and Sonar, and it is trained with the same latent diffusion objective for next-embedding prediction as in LCM’s text-only pre-training. Experiments on a large-scale multilingual and -modal instruction–tuning data mixture highlight the potential of v-LCM: v-LCM matches state-of-the-art vision-language models on tasks covering image/video captioning and question answering, while significantly outperforming them across 61 rich- to low-resource languages out of all 62 tested languages.</description>
<link>https://ai.meta.com/research/publications/unified-vision-language-modeling-via-concept-space-alignment/</link>
<guid isPermaLink="false">1257213679668978</guid>
<pubDate>Fri, 27 Feb 2026 08:00:00 GMT</pubDate>
<author>Yifu Qiu, Paul-Ambroise Duquenne, Holger Schwenk</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Human & Machine Intelligence</category>
<category>Research</category>
</item>
<item>
<title>Learning Personalized Agents from Human Feedback</title>
<description>Modern AI agents are powerful but often fail to align with the idiosyncratic, evolving preferences of individual users. Prior approaches typically rely on static datasets, either training implicit preference models on interaction history or encoding user profiles in external memory. However, these approaches struggle with new users and with preferences that change over time. We introduce Personalized Agents from Human Feedback (PAHF), a framework for continual personalization in which agents learn online from live interaction using explicit per-user memory. PAHF operationalizes a three-step loop: (1) seeking pre-action clarification to resolve ambiguity, (2) grounding actions in preferences retrieved from memory, and (3) integrating post-action feedback to update memory when preferences drift. To evaluate this capability, we develop a four-phase protocol and two benchmarks in embodied manipulation and online shopping. These benchmarks quantify an agent's ability to learn initial preferences from scratch and subsequently adapt to persona shifts. Our theoretical analysis and empirical results show that integrating explicit memory with dual feedback channels is critical: PAHF learns substantially faster and consistently outperforms both no-memory and single-channel baselines, reducing initial personalization error and enabling rapid adaptation to preference shifts.</description>
<link>https://ai.meta.com/research/publications/learning-personalized-agents-from-human-feedback/</link>
<guid isPermaLink="false">910245944695813</guid>
<pubDate>Thu, 26 Feb 2026 08:00:00 GMT</pubDate>
<author>Kaiqu Liang, Julia Kruk, Shengyi Qian, Xianjun Yang, Shengjie Bi, Shaoliang Nie, Michael Zhang, Lijuan Liu, Jaime Fernández Fisac, Shuyan Zhou, Saghar Hosseini</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Conversational AI</category>
<category>Research</category>
</item>
<item>
<title>FERRET: Framework for Expansion Reliant Red Teaming</title>
<description>We introduce a multi-faceted automated red teaming framework in which the goal is to generate multi-modal adversarial conversations that would break a target model and introduce various expansions that would result in more effective and efficient adversarial conversations. The introduced expansions include: 1. Horizontal expansion in which the goal is for the red team model to self-improve and generate more effective conversation starters that would shape a conversation. 2. Vertical expansion in which the goal is to take these conversation starters that are discovered in the horizontal expansion phase and expand them into effective multi-modal conversations and 3. Meta expansion in which the goal is for the red team model to discover more effective multi-modal attack strategies during the course of a conversation. We call our framework FERRET (Framework for Expansion Reliant Red Teaming) and compare it with various existing automated red teaming approaches. In our experiments, we demonstrate the effectiveness of FERRET in generating effective multi-modal adversarial conversations and its superior performance against existing state of the art approaches.</description>
<link>https://ai.meta.com/research/publications/ferret-framework-for-expansion-reliant-red-teaming/</link>
<guid isPermaLink="false">2396003004185464</guid>
<pubDate>Fri, 13 Feb 2026 08:00:00 GMT</pubDate>
<author>Ninareh Mehrabi, Vítor Albiero, Maya Pavlova, Joanna Bitton</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Responsible AI</category>
</item>
<item>
<title>UniT: Unified Multimodal Chain-of-Thought Test-time Scaling</title>
<description>Unified models can handle both multimodal understanding and generation within a single architecture, yet they typically operate in a single pass without iteratively refining their outputs. Many multimodal tasks, especially those involving complex spatial compositions, multiple interacting objects, or evolving instructions, require decomposing instructions, verifying intermediate results, and making iterative corrections. While test-time scaling (TTS) has demonstrated that allocating additional inference compute for iterative reasoning substantially improves language model performance, extending this paradigm to unified multimodal models remains an open challenge.
We introduce UniT, a framework for multimodal chain-of-thought test-time scaling that enables a single unified model to reason, verify, and refine across multiple rounds. UniT combines agentic data synthesis, unified model training, and flexible test-time inference to elicit cognitive behaviors including verification, subgoal decomposition, and content memory. Our key findings are: (1) unified models trained on short reasoning trajectories generalize to longer inference chains at test time; (2) sequential chain-of-thought reasoning provides a more scalable and compute-efficient TTS strategy than parallel sampling; (3) training on generation and editing trajectories improves out-of-distribution visual reasoning. These results establish multimodal test-time scaling as an effective paradigm for advancing both generation and understanding in unified models.</description>
<link>https://ai.meta.com/research/publications/unit-unified-multimodal-chain-of-thought-test-time-scaling/</link>
<guid isPermaLink="false">1477019817147720</guid>
<pubDate>Wed, 11 Feb 2026 08:00:00 GMT</pubDate>
<author>Leon Liangyu Chen, Haoyu Ma, Zhipeng Fan, Ziqi Huang, Animesh Sinha, Xiaoliang Dai, Jialiang Wang, Zecheng He, Jianwei Yang, Chunyuan Li, Junzhe Sun, Chu Wang, Serena Yeung-Levy, Felix Juefei-Xu</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Research</category>
<category>Computer Vision</category>
</item>
<item>
<title>AIRS-Bench: a Suite of Tasks for Frontier AI Research Science Agents</title>
<description>LLM agents hold significant promise for advancing scientific research. To accelerate this progress, we introduce AIRS-Bench (the AI Research Science Benchmark), a suite of 20 tasks sourced from state-of-the-art machine learning papers. These tasks span diverse domains, including language modeling, mathematics, bioinformatics, and time series forecasting. AIRS-Bench tasks assess agentic capabilities over the full research lifecycle---including idea generation, experiment analysis and iterative refinement---without providing baseline code. The AIRS-Bench task format is versatile, enabling easy integration of new tasks and rigorous comparison across different agentic frameworks. We establish baselines using frontier models paired with both sequential and parallel scaffolds. Our results show that agents exceed human SOTA in four tasks but fail to match it in sixteen others. Even when agents surpass human benchmarks, they do not reach the theoretical performance ceiling for the underlying tasks. These findings indicate that AIRS-Bench is far from saturated and offers substantial room for improvement.
We open-source the AIRS-Bench task definitions and evaluation code to catalyze further development in autonomous scientific research.</description>
<link>https://ai.meta.com/research/publications/airs-bench-a-suite-of-tasks-for-frontier-ai-research-science-agents/</link>
<guid isPermaLink="false">2685677328474642</guid>
<pubDate>Tue, 10 Feb 2026 08:00:00 GMT</pubDate>
<author>Alisia Lupidi, Bhavul Gauri, Thomas Simon Foster, Bassel Al Omari, Despoina Magka, Alberto Pepe, Alexis Audran-Reiss, Muna Aghamelu, Nicolas Baldwin, Lucia Cipolina-Kun, Jean-Christophe Gagnon-Audet, Chee Hau Leow, Sandra Lefdal, Hossam Mossalam, Abhinav Moudgil, Saba Nazir, Emanuel Tewolde, Isabel Urrego, Jordi Armengol-Estape, Amar Budhiraja, Gaurav Chaurasia, Abhishek Charnalia, Derek Dunfield, Karen Hambardzumyan, Daniel Izcovich, Martin Josifoski, Ishita Mediratta, Kelvin Niu, Parth Pathak, Michael Shvartsman, Edan Toledo, Anton Protopopov, Roberta Raileanu, Alexander Miller, Tatiana Shavrina, Jakob Foerster, Yoram Bachrach</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1mN_MhjGmwEVciqjtj4IB22ieXnbC967rn7zSmnNC67Q&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>NLP</category>
</item>
<item>
<title>Reducing Government Costs and Increasing Access to Greenspaces in the United Kingdom with DINO</title>
<description>Meta's DINOv2 model is enhancing reforestation efforts around the world. Learn how the UK government is using DINO to help reduce costs and increase access to greenspaces.</description>
<link>https://ai.meta.com/blog/forest-research-dino/</link>
<guid isPermaLink="false">1369350751606207</guid>
<pubDate>Mon, 09 Feb 2026 17:25:45 GMT</pubDate>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.2365-6/631673937_4911145895833160_8225979506389394021_n.gif?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=N94hGWEZkvUQ7kNvwFjZ8Yo&_nc_oc=Adqc7mnraWiwtLCGQIcOu8yhGrf3BlxZ7tS80eVhUCVH2KjTnoQj7hIkwIYd5VoJi50&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af0LlOIkEm29R3HGHameX19Sx_n4VVSfAA46Q8uHg--dGQ&oe=6A08CAFF" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
<category>Open Source</category>
</item>
<item>
<title>Multi-room Apartments Simulation (MRAS) Dataset | Meta AI Research</title>
<description>The Multi-Room Apartments Simulation (MRAS) dataset is a multi-modal dataset created for the task of estimating spatially-distributed acoustic parameters in complex scenes.</description>
<link>https://ai.meta.com/datasets/mras-simulated-rooms-dataset/</link>
<guid isPermaLink="false">2247220155683256</guid>
<pubDate>Wed, 28 Jan 2026 18:49:18 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/612410014_2220471548441535_606775622330495537_n.jpg?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=1bE1l45d6ZQQ7kNvwGzExIi&_nc_oc=Ado-6Ct9Xx3eWjk90FaKNEtJode5O_FsHbdvIUqlBg0pmSXdVV101WYNtSFvKg9wQLc&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1H2zBtut_FPZYOrLGUpHF9IppsZjIdP5xbafAfU_oyIQ&oe=6A08CB19" type="image/jpeg"></enclosure>
<category>DATASET</category>
</item>
<item>
<title>How DINO and SAM are Helping Modernize Essential Medical Triage Practices</title>
<description>By leveraging advanced AI models, teams at the University of Pennsylvania are aiming to bring cutting-edge automation to emergency response.</description>
<link>https://ai.meta.com/blog/upenn-dino-sam-helping-medical-triage/</link>
<guid isPermaLink="false">3255814814579669</guid>
<pubDate>Thu, 18 Dec 2025 18:00:20 GMT</pubDate>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.2365-6/597680380_857278207264023_6561532616114241333_n.png?_nc_cat=109&ccb=1-7&_nc_sid=e280be&_nc_ohc=2MKpeQyaIXMQ7kNvwEEo7tY&_nc_oc=Adpq6zSTC_WtSMZ--D1-Nop-5rrWQy13gjmc5RtnZOIiPEsKiQnSqcnrgEWUNyFyi-0&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1YDqAmgRytKyuklQlHOEBNFgCSwRRagCcwGoar5n_Vmw&oe=6A08D437" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>The Universities Space Research Association Applies Segment Anything Model for Responding to Flood Emergencies</title>
<description>The Universities Space Research Association and Meta are collaborating to help support water observing systems set up by the U.S. Geological Survey. </description>
<link>https://ai.meta.com/blog/usra-sam-flood-emergencies/</link>
<guid isPermaLink="false">1381613153671193</guid>
<pubDate>Thu, 18 Dec 2025 18:00:12 GMT</pubDate>
<enclosure url="https://scontent-hkg4-2.xx.fbcdn.net/v/t39.2365-6/600304853_1519099669397813_6606203512792368025_n.png?_nc_cat=111&ccb=1-7&_nc_sid=e280be&_nc_ohc=pBY1dQ-wuD4Q7kNvwEWzj6M&_nc_oc=AdpdwPfP0JGOaTgweBs95cQDXBToqdvNvtjoFQL4YZIyjiTJR4XYN5Ob9gFskGyfO-E&_nc_zt=14&_nc_ht=scontent-hkg4-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1VE8kVa6TXKCxeupCJEEdzIuj0rdhLeOZZ53IFvHXx-g&oe=6A08B857" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Computer Vision</category>
</item>
<item>
<title>How Orakl Oncology is using DINOv2 to accelerate cancer treatment discovery</title>
<description>Orakl Oncology aims to accelerate cancer research and drug development by combining experimental, lab-based insights and machine learning.</description>
<link>https://ai.meta.com/blog/orakl-oncology-dinov2-accelerating-cancer-treatment/</link>
<guid isPermaLink="false">1035491295071881</guid>
<pubDate>Thu, 20 Feb 2025 17:00:31 GMT</pubDate>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.2365-6/480460718_649637170770478_3994719197111548806_n.jpg?_nc_cat=106&ccb=1-7&_nc_sid=e280be&_nc_ohc=FE_ybvE6IoMQ7kNvwHPNCAd&_nc_oc=Adr7EI6QaBcpn_YJLf8HybvAzlgpRsVmByc2YK5WgnzN5icjGiL0HXeoQ6zPio2ABVw&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af1zOTRb9gox_L6KQ5hVwoxVPPXCZvOSuZLB5TYDRcexxQ&oe=6A08C327" type="image/jpeg"></enclosure>
<category>BLOG</category>
<category>Open Source</category>
</item>
<item>
<title>Aaron Defazio</title>
<description>Aaron's research focuses on improving the practice of machine learning through the development of more reliable and theoretically sound methods such as ...</description>
<link>https://ai.meta.com/people/1115638629589333/aaron-defazio/</link>
<guid isPermaLink="false">1115638629589333</guid>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.35477-6/431543048_2088159024878120_1553708304445833368_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=109&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=vd7kajaxkn4Q7kNvwEG4SoE&_nc_oc=Adrjo9WKzGduyHihaJ9oRKDhf_u31CRkHOA0HsecIfZO9KzGgWUCVqREJtVAKwUQ7Z8&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af2Mu1DrRerQ41Wk7n-SeMza_77Iyq23BttHDeyIj7QyqQ&oe=69F449C8" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Speech & Audio</category>
<category>NLP</category>
<category>Core Machine Learning</category>
<category>Theory</category>
</item>
<item>
<title>Abhimanyu Dubey</title>
<description>Abhimanyu is a Research Scientist at Facebook AI Research working on problems in machine learning. He received his Ph.D. in 2021 from the Massachusetts ...</description>
<link>https://ai.meta.com/people/1401136157273652/abhimanyu-dubey/</link>
<guid isPermaLink="false">1401136157273652</guid>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.35477-6/433945609_1344738226210370_2541979406521528116_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=54biatSobnEQ7kNvwG8A8ef&_nc_oc=AdpIQMfdVEP3_j1Uz5oXdMdPzutCxBnBHAZGm-YnIZrdUDNjyYZHYzJihUGycEfbl7U&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af2AxL8gEsdKDK3hLWQi3bIFV5GHo4E-sGTClpkOtjXGjQ&oe=69F4610D" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Computer Vision</category>
<category>Human & Machine Intelligence</category>
<category>Core Machine Learning</category>
<category>Research</category>
<category>ML Applications</category>
</item>
<item>
<title>Abhishek Das</title>
<description>Abhishek is a Research Scientist at Facebook AI Research (FAIR). His research focuses on deep learning and its applications in climate change, and in bu...</description>
<link>https://ai.meta.com/people/401335699167238/abhishek-das/</link>
<guid isPermaLink="false">401335699167238</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431537252_1543342826243118_6010155505132587328_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=102&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=7bKy_GgNvJUQ7kNvwF8DKlL&_nc_oc=AdoEjkQpXT8Ra71Iwlr5ftbfjSjVC-TbEkqvvy2zTC8X5euaGIYkMJEJO9oLdq0HAzc&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af0vxZWyhODziByPdqIUz9p3GcMwwFk6tMJR9pgtZ0d0GA&oe=69F44FDB" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Research</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
</item>
<item>
<title>Adam Polyak</title>
<description>Adam is a research engineer at Facebook AI Research (FAIR), focusing on deep learning, cross domain image generation, audio generation, and voice conver...</description>
<link>https://ai.meta.com/people/1434021433901222/adam-polyak/</link>
<guid isPermaLink="false">1434021433901222</guid>
<enclosure url="https://scontent-hkg4-2.xx.fbcdn.net/v/t39.35477-6/431518452_1124340545357794_1947310908456905622_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=111&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=qnsH_8koyHsQ7kNvwGzcNSM&_nc_oc=AdoJwpP9r3yeYpvkAnn-ViHaVGRWRZrNmZdEy7rSFN1ArYwY8oe5FRSvKafs-qATreY&_nc_zt=14&_nc_ht=scontent-hkg4-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af3SL3I_vuhN7kLZLPwCyYyw9t1kkWJP3cGM3SDjTS-DtQ&oe=69F46FBA" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Research</category>
<category>Speech & Audio</category>
<category>ML Applications</category>
</item>
<item>
<title>Adeel Cheema</title>
<description>Adeel is a software engineer working on transparency &amp; control within Meta’s Responsible AI pillar. He focuses on creating product experiences to expl...</description>
<link>https://ai.meta.com/people/1808244339639715/adeel-cheema/</link>
<guid isPermaLink="false">1808244339639715</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431551867_767017845524520_8549663744281985032_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=3dHbSn4toEQQ7kNvwF1n4d-&_nc_oc=AdqXznOtgkf_joyKpadW76WIivj8knZ7YKnKZN7RH7Q1e9XhZ_5ak48_CyT-D43hes8&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af169nMZxw-EU37ZmpyZ_E_u8oZnS5PPK4hvB4dGFK_OMQ&oe=69F44D79" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Adina Williams</title>
<description>Adina is a Research Scientist at Facebook AI Research in NYC (started October 2018). Previously, she earned her PhD at New York University in the Depart...</description>
<link>https://ai.meta.com/people/1396973444287406/adina-williams/</link>
<guid isPermaLink="false">1396973444287406</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431521840_2635057386653799_4484141047373535802_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=104&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=D9tP58vAF5kQ7kNvwF7C6fw&_nc_oc=AdpvPt7C_Jjhp4joD3b5UB-POrJZepNkHFJ0BaE0I8T4EVB_gP9eS-VRbIDEzbFKXCs&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af2CQM0Xt6o9tS_Ly49SzLpWYd8XSVeoAAQbCTx_2oCufA&oe=69F467EC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>Computer Vision</category>
<category>Core Machine Learning</category>
<category>Open Source</category>
<category>NLP</category>
<category>Conversational AI</category>
<category>Human & Machine Intelligence</category>
<category>ML Applications</category>
</item>
<item>
<title>Adriana Romero Soriano</title>
<description>Adriana is currently a research scientist at Facebook AI Research and an adjunct professor at McGill University. The goal of her research is to develop ...</description>
<link>https://ai.meta.com/people/425710613281765/adriana-romero-soriano/</link>
<guid isPermaLink="false">425710613281765</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/431577893_1361257101250742_6288541672250118710_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=nsGSoSWY8rkQ7kNvwHp-g62&_nc_oc=AdpqSlSYXprfSM2Tg780n64EZrGd82vlPOVyEfr8PRGp2ppWUVoQzwDtVilqFfNf3lI&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af0DnZFYEItGV1Tgx7ZILYceKUTMa8EUsxC4ZlS7g_g63Q&oe=69F44DCC" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Research</category>
<category>NLP</category>
<category>Computer Vision</category>
<category>Core Machine Learning</category>
<category>Conversational AI</category>
<category>Reinforcement Learni9ng</category>
</item>
<item>
<title>Akhil Mathur</title>
<description>Akhil Mathur is a Research Scientist in the Llama Research team at Meta, working on enhancing the reasoning abilities of Llama models. Previously, as a ...</description>
<link>https://ai.meta.com/people/1251205439201018/akhil-mathur/</link>
<guid isPermaLink="false">1251205439201018</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/457008091_844600247808930_1170475198447093629_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=A2xSui7yUHAQ7kNvwFS23zM&_nc_oc=AdrahVtA7Wbc1v3tY3QbRSHzsNwyMLaCTMf2PTn736NWv6hhVgubMa_4-14IsGUrHK4&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af0OLn6R38iv1VCuhgJyMVCjXU6pLDS78vOSeXYXmQox5g&oe=69F4499C" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
<item>
<title>Akshara Rai</title>
<description>Akshara is a Research Scientist at Facebook AI Research, working at the intersection of machine learning and control. Her research aims at teaching robo...</description>
<link>https://ai.meta.com/people/423502423531745/akshara-rai/</link>
<guid isPermaLink="false">423502423531745</guid>
<enclosure url="https://scontent-hkg1-1.xx.fbcdn.net/v/t39.35477-6/433923962_1381208202581063_6502596142308749692_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=101&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=0Wm1dG_MC68Q7kNvwEQP7ar&_nc_oc=Adq_deH7ji9ZB1xmu-dPrpWID0BCX9-6V1N9jAnvZL94SgSLUQTUbTggiNCwTtCsqhY&_nc_zt=14&_nc_ht=scontent-hkg1-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af3kBB_ERsqYLq7Xn4JDhA02zujtgivm8U9yjSh1NBRudw&oe=69F43F4F" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Robotics</category>
<category>Computer Vision</category>
<category>Research</category>
<category>Reinforcement Learni9ng</category>
</item>
<item>
<title>Alborz Geramifard</title>
<description>Alborz Geramifard is a senior research manager at Meta AI supporting the Conversational AI. Prior to joining Meta, he led the conversational AI team at ...</description>
<link>https://ai.meta.com/people/363544420007153/alborz-geramifard/</link>
<guid isPermaLink="false">363544420007153</guid>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.35477-6/431495520_3457483597876936_681607123433531609_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=103&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=QzBw61zVxHEQ7kNvwFlE51j&_nc_oc=Adq6JhqFdOijHej1827AxfrBteLpqdbt7M2AEiSaSYBRb0twHVO0Vkzc7La_KpTtqrw&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af3WxPS7fsYSoKr9gF_KTUmhTOs7gKj5NJASFgu5-2Rxag&oe=69F45A51" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Robotics</category>
<category>Reinforcement Learni9ng</category>
<category>Conversational AI</category>
<category>NLP</category>
<category>Computer Vision</category>
</item>
<item>
<title>Alexander H. Miller</title>
<description>Alexander has worked in multi-domain research engineering management, supporting research engineers in London, Tel Aviv, and New York. He is advanced in...</description>
<link>https://ai.meta.com/people/264624513377272/alexander-h-miller/</link>
<guid isPermaLink="false">264624513377272</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/433950826_944833516851120_841662693953017766_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=110&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=6ftFVCshb-8Q7kNvwEmShVr&_nc_oc=AdqDT05sTcWqiXL8CooTGILItRGi0hr8Ylblc4YGYl0M5hVqjpjk1UwnAs5-tmRvHo8&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af2c5Hsyn-KW69TMWmodEAjSI-UcVySfdKLj37-nE0FyNw&oe=69F45C6C" type="image/jpeg"></enclosure>
<category>PERSON</category>
<category>Reinforcement Learni9ng</category>
<category>NLP</category>
<category>Graphics</category>
</item>
<item>
<title>Amey Porobo Dharwadker</title>
<description>Amey Porobo Dharwadker works as a Machine Learning Engineering Manager, leading Facebook's Video Recommendations Ranking team. Renowned for his expertis...</description>
<link>https://ai.meta.com/people/393924029921120/amey-porobo-dharwadker/</link>
<guid isPermaLink="false">393924029921120</guid>
<enclosure url="https://scontent-hkg4-1.xx.fbcdn.net/v/t39.35477-6/431510720_7490594577724788_4289788014840147933_n.jpg?stp=dst-jpg_s280x280_tt6&_nc_cat=106&ccb=1-7&_nc_sid=2b8f6b&_nc_ohc=CDFvi0iDZRcQ7kNvwHngb7S&_nc_oc=Ado-on7H7BL3gsoNI2IvWjApO7MmYjUwGlBV_K3EOEWlY71TKmdUsBi5V96q7WVtg28&_nc_zt=14&_nc_ht=scontent-hkg4-1.xx&_nc_gid=tFwTmxnYK7lI0uLBZdcVLw&_nc_ss=73289&oh=00_Af3OSvSyJTaZrd4DTlne1y0x5LV6ihNmtsCE_sXbbAeDRg&oe=69F440F5" type="image/jpeg"></enclosure>
<category>PERSON</category>
</item>
</channel>
</rss>http://localhost:1200/meta/ai/global-search/q=llama%26content_types=publication%26years=2024,2025%26sort_by=MOST_RECENT - Success ✔️<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Meta AI Global Search — q=llama · content_types=publication · years=2024,2025</title>
<link>https://ai.meta.com/global_search/</link>
<atom:link href="http://localhost:1200/meta/ai/global-search/q=llama%26content_types=publication%26years=2024,2025%26sort_by=MOST_RECENT" rel="self" type="application/rss+xml"></atom:link>
<description>Search results from ai.meta.com/global_search/. - Powered by RSSHub</description>
<generator>RSSHub</generator>
<webMaster>contact@rsshub.app (RSSHub)</webMaster>
<language>en</language>
<lastBuildDate>Sun, 26 Apr 2026 21:22:53 GMT</lastBuildDate>
<ttl>5</ttl>
<item>
<title>Accelerating a Triton Fused Kernel for W4A16 Quantized Inference with SplitK Work Decomposition</title>
<description>We propose an implementation of an efficient fused matrix multiplication kernel for W4A16 quantized inference, where we perform dequantization and GEMM in a fused kernel using a SplitK work decomposition.
Our implementation shows improvement for the type of skinny matrix-matrix multiplications found in foundation model inference workloads. In particular, this paper surveys the type of matrix multiplication between a skinny activation matrix and a square weight matrix. Our results show an average of 65\% speed improvement on A100, and an average of 124\% speed improvement on H100 (with a peak of 295\%) for a range of matrix dimensions including those found in a llama-style model, where m &lt; n = k.</description>
<link>https://ai.meta.com/research/publications/accelerating-a-triton-fused-kernel-for-w4a16-quantized-inference-with-splitk-work-decomposition/</link>
<guid isPermaLink="false">347040654790634</guid>
<pubDate>Tue, 09 Jan 2024 16:00:00 GMT</pubDate>
<author>Less Wright, Adnan Hoque</author>
<enclosure url="https://scontent-hkg1-2.xx.fbcdn.net/v/t39.2365-6/91980731_255741849160254_7676025225686810624_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=XbaweBXNr3EQ7kNvwE_d56M&_nc_oc=Adqw0wJ-lcgbHiy5i0Qgz-s2DrxwaCErW-0srwyfmX5vRR5l-s_OXydZbz72pNgWnMA&_nc_zt=14&_nc_ht=scontent-hkg1-2.xx&_nc_gid=u3t2r3PqikOgRFQsNlWawQ&_nc_ss=73289&oh=00_Af0MkMUQl_zJugDxgaPVv8IMAZhsqo5m4gC5eS4bUeYhYg&oe=6A08C32C" type="image/jpeg"></enclosure>
<category>PUBLICATION</category>
<category>Core Machine Learning</category>
</item>
</channel>
</rss>... |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Involved Issue / 该 PR 相关 Issue
None
Example for the Proposed Route(s) / 路由地址示例
New RSS Route Checklist / 新 RSS 路由检查表
/meta/ai/blog); GraphQL request reuses the LSD token. ofetch's existing retry-on-400 withx-prefer-proxy: 1covers Meta's edge throttling.published_time→parseDatems)PuppeteerNote / 说明
Adds support for
https://ai.meta.com/global_search/, which is backed by theuseFBAIGlobalSearchQueryGraphQL query (doc_id=9716930201759979).Filters (path-encoded
routeParams):q— search querycontent_types— comma-separated:person,publication,blog,dataset,event,toolresearch_areas— e.g.natural-language-processing,computer-visionfilter_tags—research,ml-applications,open-source,developer-tools,ar-vr,hardwareyears— e.g.2024,2025location_cities— publication venues, e.g.AAAI,ACLalphabetical_filter— single letter (pairs withcontent_types=person+sort_by=ALPHABETICAL)sort_by—RELEVANCE(default) /MOST_RECENT/ALPHABETICAL/RANDOMoffset— pagination offsetlimitstays as a regular query string (?limit=N) because the cache layer keys on it. Path-encodedrouteParamskeeps each filter combination in its own cache entry, matching the convention used byweibo/keyword/:keyword/:routeParams?. The full set ofsort_by/content_typesenum values was discovered empirically by probing the GraphQL endpoint.The LSD/SiteData fetch and GraphQL body builder were extracted into
lib/routes/meta/utils.tssoai-blog.tsandai-global-search.tsshare the boilerplate./meta/ai/bloghas been refactored onto the helper and regression-tested locally — output is unchanged.To combine multiple filters in one URL, encode
&as%26inside the path (e.g.q=llama%26content_types=publication).