PyGeNNProject — Macaque Visual Cortex Large-Scale Spiking Neural Network Simulation Platform

A large-scale Spiking Neural Network (SNN) simulation platform based on PyGeNN (GPU-enhanced Neural Networks) + Numba CUDA + multiprocessing/MPI parallelism, designed for simulating neuronal population activity in the macaque visual cortex.

---

Overview

This project constructs large-scale SNN models ranging from single neurons and single cortical columns to multi-area full visual cortex, covering 32 visual cortical areas with multi-GPU parallel acceleration. The models are based on the macaque visual cortex microcircuit framework by Schmidt et al. and implemented via PyGeNN for efficient GPU simulation.

Model Directory

Directory	Description
SingleNeuron/	Single-neuron LIF model tests
expLIF/	Exponential LIF neuron model definition (with adaptive exponential)
DualEXP/	Dual Exponential Synapse model
IzhkNeuron/	Izhikevich neuron model implementation
EIBalance/	Excitatory-inhibitory balanced network model
SingleColumn/	Single cortical column model (Potjans-Diesmann style, 16 neuron populations)
DoubleColumn/	Dual cortical column interconnection model
MultiColumn/	Multi-column extension model
MultiLayer/	Multi-layer network model
potjansModel/	Potjans-Diesmann 2014 cortical microcircuit reproduction
SchmidtModel/	Schmidt macaque visual cortex microcircuit model
CustomModel/	Custom full-cortex connectivity model (supports 32 brain areas)
HMAM/	Human Multi-Area Model
HMAM_MPI/	MPI distributed multi-node version of HMAM
MAM_MPI/	Main model: Multi-Area MPI parallel model (32 areas, multi-GPU multi-process)
ScalingTest/	Model scalability tests
ProjectOfStimExcNeuronInFirstArea/	Stimulation experiment on excitatory neurons in the first area
model_info_schmidt_motif_diff_s/	Connectivity data files for Schmidt motif model

Technology Stack

Layer	Technology
Neuron Models	LIF / Exponential LIF / Izhikevich
Synapse Models	ExpCurr (exponential decay current) / Dual Exponential
GPU Acceleration	PyGeNN + Numba CUDA (custom kernels)
Parallel Frameworks	Python multiprocessing (multi-GPU) + MPI (multi-node)
Data Processing	NumPy, SciPy, Pandas
Visualization	Matplotlib, Seaborn
Build System	setuptools

Main Model: MAM_MPI

MAM_MPI is the most comprehensive simulation model in this project, simulating 32 macaque visual cortical areas (V1, V2, V3, V4, MT, MSTd, LIP, FEF, 7a, 46, AITd, etc.), with each area containing 5 cortical layers and 4–5 neuron subtypes.

# Basic run
python MAM_MPI/CustomModel_MPI.py --duration 1000 --AreaNum 32 --scale 1.0

# Run with stimulation
python MAM_MPI/CustomModel_MPI.py --duration 1000 --AreaNum 32 --scale 1.0 --stim --stim-start 300 --stim-end 800

See MAM_MPI/readme.md for details.

Dependencies

• Python 3.8+
• PyGeNN (GPU SNN simulation framework)
• Numba (CUDA JIT compilation)
• NumPy, SciPy, Pandas, Matplotlib
• nested_dict, pynvml

pip install pandas matplotlib scipy seaborn numba nested_dict pynvml

Installation

git clone git@github.com:OOOOOmoi/PyGeNNProject.git
cd PyGeNNProject
pip install -e .

Data Files

• custom_Data_Model_3396.json — Complete connection parameters for 32 brain areas (3396 mm² cortical surface area)
• default_Data_Model__*.json — Model parameters for different cortical surface areas (50/70/100 mm²)
• indegrees_full.json / outdegrees_full.json — Whole-brain in-degree and out-degree data
• viscortex_raw_data.json — Visual cortex raw connectivity data
• Fac_result.json — Facilitation factor results

References

• Schmidt et al., "A multi-scale layer-resolved spiking network model of resting-state dynamics in macaque visual cortical areas", PLOS Computational Biology, 2018.
• Potjans & Diesmann, "The cell-type specific cortical microcircuit: relating structure and activity in a full-scale spiking network model", Cerebral Cortex, 2014.
• Knight & Nowotny, "GPUs outperform current HPC and neuromorphic solutions in terms of speed and energy when simulating a highly-connected cortical model", Frontiers in Neuroscience, 2018.