Training Full Spike Neural Networks via Auxiliary Accumulation Pathway

Introduction

This repository contains the official PyTorch implementation of the following paper:

Training Full Spike Neural Networks via Auxiliary Accumulation Pathway,
Guangyao Chen, Peixi Peng, Guoqi Li, Yonghong Tian [arXiv][Bibtex]

Updates

[05/2023] Code are released.

Dependency

We suggest to use anaconda install all packages.

Install torch>=1.5.0 by referring to:

https://pytorch.org/get-started/previous-versions/

Install tensorboard:

pip install tensorboard

The origin codes uses a specific SpikingJelly. To maximize reproducibility, the user can download the latest SpikingJelly and rollback to the version that we used to train:

git clone https://github.com/fangwei123456/spikingjelly.git
cd spikingjelly
git reset --hard 2958519df84ad77c316c6e6fbfac96fb2e5f59a3
python setup.py install

Running Examples

Train on ImageNet

cd imagenet

Train the FSNN-18 (AAP) with 8 GPUs:

python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --cos_lr_T 320 --model dsnn18 -b 64 --output-dir ./logs --tb --print-freq 500 --amp --cache-dataset --connect_f IAND --T 4 --lr 0.1 --epoch 320 --data-path /datasets/imagenet

Citation

If you find our work useful for your research, please consider giving a star ⭐ and citation 🍺:

@article{chen2023training,
  title={Training Full Spike Neural Networks via Auxiliary Accumulation Pathway},
  author={Chen, Guangyao and Peng, Peixi and Li, Guoqi and Tian, Yonghong},
  journal={arXiv preprint arXiv:2301.11929},
  year={2023}
}

Acknowledgement

This code is built using the spikingjelly framework, the syops-counter tool and the Spike-Element-Wise-ResNet repository.

iCGY96/AAP