Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion docs/assets/mermaid/.cache-key
Original file line number Diff line number Diff line change
@@ -1 +1 @@
866dfc20ff03579b4ed9fe88ba6b151b5790e9c1c1aec829d93657714545660f
bce837d901f7461a545c93ea2ad71c7b1a4a51de72d4255a4d38ff7831336d06
2 changes: 1 addition & 1 deletion docs/assets/mermaid/chat-products_mermaid_0_d921ced9.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 1 addition & 1 deletion docs/assets/mermaid/failure-cases_mermaid_0_6cb86e51.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 1 addition & 1 deletion docs/assets/mermaid/index_mermaid_0_4bfd489a.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 1 addition & 1 deletion docs/assets/mermaid/index_mermaid_1_4c49cd69.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 1 addition & 1 deletion docs/assets/mermaid/prompt-basic_mermaid_0_72623ff4.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 1 addition & 1 deletion docs/assets/mermaid/prompt-basic_mermaid_1_5d4e2cd9.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 1 addition & 1 deletion docs/assets/mermaid/prompt-basic_mermaid_2_4fa92616.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 1 addition & 1 deletion docs/assets/mermaid/stable-output_mermaid_0_a4883ca0.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 1 addition & 1 deletion docs/assets/mermaid/stable-output_mermaid_1_b549843d.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 1 addition & 1 deletion docs/assets/mermaid/structure_mermaid_0_39cfc0b3.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 1 addition & 1 deletion docs/assets/mermaid/tasks_mermaid_0_f5e00efb.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 1 addition & 1 deletion docs/assets/mermaid/templates_mermaid_0_bf562375.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
17 changes: 15 additions & 2 deletions docs/basics/deep-learning.md
Original file line number Diff line number Diff line change
Expand Up @@ -359,8 +359,21 @@

## 延伸阅读

- [什么是机器学习](machine-learning.md) —— 回顾机器学习的大框架,理解深度学习在其中的位置
- [什么是 LLM](what-is-llm.md) —— 了解深度学习在语言方向上的最新产物:大语言模型
- [什么是机器学习](../machine-learning/) —— 回顾机器学习的大框架,理解深度学习在其中的位置
- [什么是 LLM](../what-is-llm/) —— 了解深度学习在语言方向上的最新产物:大语言模型

## 下一步

<div markdown="1" style="border:1px solid var(--md-default-fg-color--lightest);border-left:4px solid var(--md-accent-fg-color);border-radius:0.85rem;padding:1rem 1.1rem;margin:0.9rem 0;background:linear-gradient(135deg,var(--md-code-bg-color),rgba(233,30,99,0.06));">

理解了深度学习之后,下一站建议看:

<a href="../what-is-llm/" style="display:block;margin-top:0.75rem;padding:0.85rem 1rem;border-radius:0.65rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
<strong>什么是 LLM →</strong><br>
<span style="color:var(--md-default-fg-color--light);font-size:0.92rem;">了解深度学习在语言方向上的关键产物——大语言模型如何生成文字、接住上下文。</span>
</a>

</div>

## 练习题 / 小实验

Expand All @@ -370,17 +383,17 @@

??? question "参考答案"
**思路 1:从数据的结构化程度看**

Check warning on line 386 in docs/basics/deep-learning.md

View workflow job for this annotation

GitHub Actions / build

markdown/trailing-whitespace

行尾存在多余空白。 建议:删除行尾空格;确需 Markdown 强制换行时保留两个空格。
图像和语言是典型的**非结构化数据**——像素和文字没有固定的格式,无法用表格的行列直接表达。深度学习恰恰擅长从非结构化数据里自动提炼特征,而传统机器学习需要人先把数据整理成结构化格式(特征工程)。表格数据本身就是结构化的,两者的优势正好反转。

Check warning on line 388 in docs/basics/deep-learning.md

View workflow job for this annotation

GitHub Actions / build

markdown/trailing-whitespace

行尾存在多余空白。 建议:删除行尾空格;确需 Markdown 强制换行时保留两个空格。
**思路 2:从数据量看**

Check warning on line 390 in docs/basics/deep-learning.md

View workflow job for this annotation

GitHub Actions / build

markdown/trailing-whitespace

行尾存在多余空白。 建议:删除行尾空格;确需 Markdown 强制换行时保留两个空格。
深度学习需要海量数据才能发挥威力。互联网上的图片和文本唾手可得,但企业的销售记录、生产数据往往只有几千到几万条——根本不够深度学习「吃饱」。数据量不足时,深度学习容易过拟合,传统机器学习反而更稳健。

Check warning on line 392 in docs/basics/deep-learning.md

View workflow job for this annotation

GitHub Actions / build

markdown/trailing-whitespace

行尾存在多余空白。 建议:删除行尾空格;确需 Markdown 强制换行时保留两个空格。
**思路 3:从特征的层次性看**

Check warning on line 394 in docs/basics/deep-learning.md

View workflow job for this annotation

GitHub Actions / build

markdown/trailing-whitespace

行尾存在多余空白。 建议:删除行尾空格;确需 Markdown 强制换行时保留两个空格。
图像和语言有天然的层次结构(像素→边缘→部件→概念;字→词→句→段),深度神经网络能很好地捕捉这种层次。表格数据的特征之间往往没有这种嵌套的层次关系,用简单的线性模型或树模型就足够了,强行用深度学习反而增加复杂度。

Check warning on line 396 in docs/basics/deep-learning.md

View workflow job for this annotation

GitHub Actions / build

markdown/trailing-whitespace

行尾存在多余空白。 建议:删除行尾空格;确需 Markdown 强制换行时保留两个空格。
**综合结论**:图像和语言的本质是非结构化、海量数据、有层次结构的;表格数据是结构化、小数据、无层次结构的。深度学习擅长的,恰好是传统机器学习不擅长的,所以两者在不同场景下各领风骚。

---
Expand All @@ -389,13 +402,13 @@

??? question "参考答案"
**要解决的核心问题:梯度消失**

Check warning on line 405 in docs/basics/deep-learning.md

View workflow job for this annotation

GitHub Actions / build

markdown/trailing-whitespace

行尾存在多余空白。 建议:删除行尾空格;确需 Markdown 强制换行时保留两个空格。
反向传播时,梯度从输出层往回传,每经过一层就乘以一个系数(激活函数的导数)。层数多了之后,如果每层的系数都小于 1,梯度会指数级衰减到接近零,导致前面几层根本学不到东西。

Check warning on line 407 in docs/basics/deep-learning.md

View workflow job for this annotation

GitHub Actions / build

markdown/trailing-whitespace

行尾存在多余空白。 建议:删除行尾空格;确需 Markdown 强制换行时保留两个空格。
**残差连接的工作方式**

Check warning on line 409 in docs/basics/deep-learning.md

View workflow job for this annotation

GitHub Actions / build

markdown/trailing-whitespace

行尾存在多余空白。 建议:删除行尾空格;确需 Markdown 强制换行时保留两个空格。
正常情况下,每一层的输入 x 会通过权重层变成 Wx,然后经过激活函数输出 f(Wx)。

Check warning on line 411 in docs/basics/deep-learning.md

View workflow job for this annotation

GitHub Actions / build

markdown/trailing-whitespace

行尾存在多余空白。 建议:删除行尾空格;确需 Markdown 强制换行时保留两个空格。
残差连接在此基础上加了一条「捷径」:直接让输入 x 绕过权重层,和激活后的输出加在一起:`输出 = f(Wx) + x`。

这条捷径让梯度可以直接流回去——求导时 `d/dx (f(Wx) + x) = f'(Wx)·W + 1`,那个 +1 保证梯度至少是 1,不会消失。就好像在一条拥堵的路上开了一条应急车道,让信息畅通无阻。
Expand Down
401 changes: 373 additions & 28 deletions docs/basics/hallucination.md

Large diffs are not rendered by default.

264 changes: 218 additions & 46 deletions docs/basics/index.md

Large diffs are not rendered by default.

6 changes: 3 additions & 3 deletions docs/basics/machine-learning.md
Original file line number Diff line number Diff line change
Expand Up @@ -513,10 +513,10 @@ flowchart LR
## 延伸阅读

<div markdown="1" style="display:grid;grid-template-columns:repeat(auto-fit,minmax(14rem,1fr));gap:0.85rem;margin:1rem 0;">
<a href="what-is-ai.md" style="display:block;padding:0.95rem 1rem;border-radius:0.8rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
<a href="../what-is-ai/" style="display:block;padding:0.95rem 1rem;border-radius:0.8rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
<strong>什么是 AI</strong><br><span style="color:var(--md-default-fg-color--light);font-size:0.9rem;">从更大的视角理解 AI、机器学习、深度学习的关系。</span>
</a>
<a href="deep-learning.md" style="display:block;padding:0.95rem 1rem;border-radius:0.8rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
<a href="../deep-learning/" style="display:block;padding:0.95rem 1rem;border-radius:0.8rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
<strong>什么是深度学习</strong><br><span style="color:var(--md-default-fg-color--light);font-size:0.9rem;">了解机器学习中最热门的一条技术路线。</span>
</a>
<a href="https://developers.google.com/machine-learning/crash-course/" style="display:block;padding:0.95rem 1rem;border-radius:0.8rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
Expand All @@ -533,7 +533,7 @@ flowchart LR

分清了机器学习的概念之后,下一站建议看:

<a href="deep-learning.md" style="display:block;margin-top:0.75rem;padding:0.85rem 1rem;border-radius:0.65rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
<a href="../deep-learning/" style="display:block;margin-top:0.75rem;padding:0.85rem 1rem;border-radius:0.65rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
<strong>深度学习 →</strong><br>
<span style="color:var(--md-default-fg-color--light);font-size:0.92rem;">了解机器学习中最热门的一条技术路线——用多层神经网络处理复杂任务。</span>
</a>
Expand Down
2 changes: 1 addition & 1 deletion docs/basics/model-data-training.md
Original file line number Diff line number Diff line change
Expand Up @@ -524,7 +524,7 @@ flowchart LR

理解了模型、数据、训练和推理之后,下一站建议看:

<a href="token-embedding-context.md" style="display:block;margin-top:0.75rem;padding:0.85rem 1rem;border-radius:0.65rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
<a href="../token-embedding-context/" style="display:block;margin-top:0.75rem;padding:0.85rem 1rem;border-radius:0.65rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
<strong>Token、Embedding 与上下文窗口 →</strong><br>
<span style="color:var(--md-default-fg-color--light);font-size:0.92rem;">继续拆开模型输入输出的基本单位,理解 token、向量和上下文窗口为什么影响价格、速度和效果。</span>
</a>
Expand Down
4 changes: 2 additions & 2 deletions docs/basics/multimodal-ai.md
Original file line number Diff line number Diff line change
Expand Up @@ -101,7 +101,7 @@ flowchart LR

文本是目前最成熟、最常见的 AI 输入。

你在聊天框里输入一句话,模型会先把它切成 Token,再通过 Embedding 变成向量。这个过程在 [Token、Embedding 与上下文窗口](token-embedding-context.md) 里会详细讲。
你在聊天框里输入一句话,模型会先把它切成 Token,再通过 Embedding 变成向量。这个过程在 [Token、Embedding 与上下文窗口](../token-embedding-context/) 里会详细讲。

文本 AI 擅长:

Expand Down Expand Up @@ -450,7 +450,7 @@ RAG 会在后面单独讲。这里先记住一个简单解释:RAG 就是先从

理解了多模态 AI 之后,下一站建议看:

<a href="model-data-training.md" style="display:block;margin-top:0.75rem;padding:0.85rem 1rem;border-radius:0.65rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
<a href="../model-data-training/" style="display:block;margin-top:0.75rem;padding:0.85rem 1rem;border-radius:0.65rem;background:var(--md-default-bg-color);text-decoration:none;border:1px solid var(--md-default-fg-color--lightest);">
<strong>模型、数据、训练与推理 →</strong><br>
<span style="color:var(--md-default-fg-color--light);font-size:0.92rem;">了解模型能力是怎么从数据、训练和推理流程里长出来的。</span>
</a>
Expand Down
Loading
Loading