dzy-cxy / snn_optimal_conversion_pipeline

Optimal Conversion of Conventional Artificial Neural Networks to Spiking Neural Networks

snn_optimal_conversion_pipeline

Optimal Conversion of Conventional Artificial Neural Networks to Spiking Neural Networks

Training and simulation

We suggest using file 'main_train.py' for training and file 'main_simulation.py' for simulation.

• The training and simulation parameters are collected in 'models/settings.py'.

Files

• 'main_train.py' : main training file.
• 'main_simulation.py' : main simulation file.
• 'models/settings.py' : collection of the parameters.
• 'models/spiking_layer.py' : SPIKE_layer to replace ANN's convolution layer and linear layer.
• 'models/new_relu.py' : threshold ReLU file

Pre-trained models

• All the pre-trained models we used are avilable here

Issues

• Consider the generalization, when T is large, the loss and accuracy of SNN may both decrease.

Citation

If our code is helpful to you, please cite the following paper.

@inproceedings{
deng2021optimal,
title={Optimal Conversion of Conventional Artificial Neural Networks to Spiking Neural Networks},
author={Shikuang Deng and Shi Gu},
booktitle={International Conference on Learning Representations},
year={2021},
url={https://openreview.net/forum?id=FZ1oTwcXchK}
}