Compressed Latent Replays for Lightweight Continual Learning on Spiking Neural Networks

TL;DR

This paper presents a memory-efficient Latent Replay method for Spiking Neural Networks, enabling effective continual learning on resource-constrained devices with minimal forgetting and significant accuracy.

Contribution

It introduces the first memory-efficient Latent Replay approach for SNNs, combining compression techniques to reduce memory use while maintaining high accuracy in continual learning.

Findings

Achieved 92.5% Top-1 accuracy on Heidelberg SHD dataset for Sample-Incremental tasks.

Reduced Latent Replay memory by two orders of magnitude with only 4% accuracy drop.

Learned 10 new classes with 78.4% accuracy in Multi-Class-Incremental tasks.

Abstract

Rehearsal-based Continual Learning (CL) has been intensely investigated in Deep Neural Networks (DNNs). However, its application in Spiking Neural Networks (SNNs) has not been explored in depth. In this paper we introduce the first memory-efficient implementation of Latent Replay (LR)-based CL for SNNs, designed to seamlessly integrate with resource-constrained devices. LRs combine new samples with latent representations of previously learned data, to mitigate forgetting. Experiments on the Heidelberg SHD dataset with Sample and Class-Incremental tasks reach a Top-1 accuracy of 92.5% and 92%, respectively, without forgetting the previously learned information. Furthermore, we minimize the LRs' requirements by applying a time-domain compression, reducing by two orders of magnitude their memory requirement, with respect to a naive rehearsal setup, with a maximum accuracy drop of 4%. On a Multi-Class-Incremental task, our SNN learns 10 new classes from an initial set of 10, reaching a Top-1 accuracy of 78.4% on the full test set.