TM-SNN: Threshold Modulated Spiking Neural Network for Multi-task Learning

This repository contains code implementation for the paper "TM-SNN: Threshold Modulated Spiking Neural Network for Multi-task Learning". The code uses threshold modulation to implement spiking neural networks that change their behavior. TM-SNN is tested for solving multitask classification on the NMNIST neuromorphic dataset. The results show that TM-SNN can actually learn different tasks through modifying its dynamics via modulation of the neurons’ firing threshold.

Installation

Prerequisites

The code runs in Python3.9 using Intel's Lava neuromorphic framework and the Lava-DL package. They can be found at https://github.com/lava-nc/lava and https://github.com/lava-nc/lava-dl.

Clone

Clone this repo to your local machine using https://github.com/PaoloGCD/MultiTask-SNN.git

Dataset

The experiments are performed on the NMNIST dataset. Download and extract the dataset to '.data/NMNIST' folder from https://www.garrickorchard.com/datasets/n-mnist.

Running the tests

To run the SNN for single task classification (base-case), execute:

$ sh ./experiments/NMNIST-base-case.sh

To run the SNN for two-task classification using threshold control, execute:

$ sh ./experiments/NMNIST-threshold-two-blocks.sh

To run the SNN for two-task classification using threshold control whit auxiliary block, execute:

$ sh ./experiments/NMNIST-threshold-three-blocks.sh

Authors

• Paolo G. Cachi - Virginia Commonwealth University - USA