Tensor Decomposition Based Attention Module for Spiking Neural Networks

Haoyu Deng^a, Ruijie Zhu^b, Xuerui Qiu^a, Yule Duanu^a, Malu Zhang^a,† and Liang-Jian Deng^a,†

^aUniversity of Electronic Science and Technology of China, 611731, China

^bUniversity of California, Santa Cruz, 95064, United States

^†Corresponding authors

If you have any questions, feel free to raise an issue or send a mail to academic@hydeng.cn. I will respond to you as soon as possible.

This is the official repository for paper Tensor Decomposition Based Attention Module for Spiking Neural Networks

paper:[pdf]

CIFAR10/CIFAR100

How to Run

1. Prepare environment

A Docker environment is strongly recommended, but you can also use pip to prepare the environment. If you are not familiar with Docker, you can refer to the following link to know about docker:

English Tutorial

中文教程

1) Docker file

A docker file is provided in env directory. You can build the docker image and run the container with the following commands:

docker build -t pfa ./env

2) Docker image

We provide pre-build docker image on docker hub. You can pull the image with the following command:

docker pull risingentropy409/pfa

The environment is ready to use after you have the image.

3) pip

Use the following command to setup environment with pip:

pip install -r requirements.txt

<font color="red">NOTE: If your cuda version is above 12.0, you may modify cupy-cuda11x in requirements.txt to cupy-cuda12x</font>

2.Run the code

Modify CIFAR/main.py according to your needs and run:

python main.py

FMNIST

This code is modified from the example code of the spikingjelly framework. We appreciate the authors' great contribution.

How to run

Just run:

python run.py

For more configuration, please refer to the code. The code is simple and clear.

Generation Tasks (FSVAE)

We conduct generation tasks based on FSVAE. The modified code is in the Generation folder. To run it, please refer to Generation/README.md

Citation

If you think our work is useful, please give us a warm citation!

@article{DENG2024111780,
title = {Tensor decomposition based attention module for spiking neural networks},
journal = {Knowledge-Based Systems},
volume = {295},
pages = {111780},
year = {2024},
issn = {0950-7051},
doi = {https://doi.org/10.1016/j.knosys.2024.111780},
url = {https://www.sciencedirect.com/science/article/pii/S0950705124004143},
author = {Haoyu Deng and Ruijie Zhu and Xuerui Qiu and Yule Duan and Malu Zhang and Liang-Jian Deng},
keywords = {Spiking neural network, Attention mechanism, Tensor decomposition, Neuromorphic computing}
}