Temporal Knowledge Sharing enable Spiking Neural Network Learning from Past and Future

This repository contains code from our paper Temporal Knowledge Sharing enable Spiking Neural Network Learning from Past and Future.

Requirments

braincog

numpy

scipy

pytorch >= 1.7.0

torchvision

torchmetrics==0.10.3

timm==0.6.13

tenserboard

tonic

Run

TKS is based on Brain-Cog. To Run this code, download braincog

train

python train_main.py --model metarightsltet --learner VGG_SNN -b 128 --epochs 600 --device 0 --dataset cifar10 --num-classes 101 --T 3 --step 4 --alpha 0.7 --layer-by-layer

eval

python train_main.py --model metarightsltet --learner VGG_SNN -b 128 --epochs 600 --device 0 --dataset cifar10 --num-classes 101 --T 3 --step 4 --alpha 0.7 --layer-by-layer --eval --eval [your checkpoint path]

Citation

If you use this code in your work, please cite the following paper, please cite it using

  @article{dong2023temporal,
    title={Temporal Knowledge Sharing enable Spiking Neural Network Learning from Past and Future},
    author={Dong, Yiting and Zhao, Dongcheng and Zeng, Yi},
    journal={arXiv preprint arXiv:2304.06540},
    year={2023}
  }