Continuous learning of spiking networks trained with local rules

TL;DR

This paper investigates the susceptibility of spiking neural networks to catastrophic forgetting and introduces biologically inspired local learning rules and synapse importance methods to mitigate it, demonstrating their effectiveness on standard datasets.

Contribution

The study develops a novel synapse importance method based on Langevin dynamics for SNNs trained with local rules, addressing catastrophic forgetting without global gradient-based methods.

Findings

SNNs show resilience to catastrophic forgetting with local learning rules.

The Langevin dynamics-based importance method effectively mitigates forgetting.

Experimental results on standard datasets validate the proposed approach.

Abstract

Artificial neural networks (ANNs) experience catastrophic forgetting (CF) during sequential learning. In contrast, the brain can learn continuously without any signs of catastrophic forgetting. Spiking neural networks (SNNs) are the next generation of ANNs with many features borrowed from biological neural networks. Thus, SNNs potentially promise better resilience to CF. In this paper, we study the susceptibility of SNNs to CF and test several biologically inspired methods for mitigating catastrophic forgetting. SNNs are trained with biologically plausible local training rules based on spike-timing-dependent plasticity (STDP). Local training prohibits the direct use of CF prevention methods based on gradients of a global loss function. We developed and tested the method to determine the importance of synapses (weights) based on stochastic Langevin dynamics without the need for the gradients. Several other methods of catastrophic forgetting prevention adapted from analog neural networks were tested as well. The experiments were performed on freely available datasets in the SpykeTorch environment.