Towards Biological Continual Learning with Spiking Hopfield Networks

Matthew Lai, Longbing Cao

Keywords: Spike-Time Dependent Plasticity, SNN, Associative Memory, Modern Hopfield Network, Catastrophic Forgetting, Elastic Weight Consolidation

Abstract: Modern Hopfield networks are often viewed as biologically inspired associative memories, yet they lack the spiking dynamics and local learning rules that underpin real neural computation. In this work, we introduce a Spiking Hopfield Network (SHN) that incorporates discrete spike-based communication and a spike-timing–dependent plasticity (STDP) rule, enhancing biological plausibility while retaining the network’s capacity for online learning. To further support continual updates, we propose an Elastic Weight Consolidation (EWC)–inspired mechanism adapted to this local learning setting, reducing catastrophic forgetting. Together, these contributions yield a lightweight and biologically grounded framework that combines efficient memory retrieval with resilience to continual adaptation.

Primary Area: other topics in machine learning (i.e., none of the above)

Submission Number: 3798