CSNN & VDSP

This is the code for the paper Unsupervised and efficient learning in sparsely activated convolutional spiking neural networks enabled by voltage-dependent synaptic plasticity, available here.

Getting started

The code was written in Python 3.9.

Dependencies

• tqdm (4.64.0)
• numpy (1.20.1)
• librosa (0.9.2)
• scikit_learn (1.1.2)
• torch (1.12.1)
• python-mnist (0.7)

Installation

Dependencies for each subdirectory can be installed with :

pip3 install -r requirements.txt

Also, the library libsndfile must be installed. (dnf install libsndfile on fedora, apt install libsndfile1 on ubuntu).

Usage

To train and evaluate the CSNN on the MNIST or the TIDIGITS task, run the following command in the desired subdirectory :

python3 snn.py

Several other functions are available in the file snn.py (and also snn_analysis.py for MNIST) to reproduce experiments presented in the paper.

Acknowledgements

• Institut Interdisciplinaire d'Innovation Technologique (3IT), Université de Sherbrooke.
• Laboratoire Nanotechnologies Nanosystèmes (LN2) – CNRS UMI-3463, Université de Sherbrooke, Sherbrooke, Canada
• Institute of Electronics, Microelectronics and Nanotechnology (IEMN), Université de Lille, Villeneuve d’Ascq, France.
• NECOTIS Research Lab, Electrical and Computer Engineering Dep., Université de Sherbrooke, Sherbrooke, Canada.