Evolutionary Spiking Neural Networks: A Survey

TL;DR

The paper addresses the challenge of designing and training spiking neural networks (SNNs) by surveying how evolutionary algorithms can optimize architectures, neuron dynamics, and learning rules across NAS and non-NAS approaches. It documents methods like AutoSNN, SNASNet, SpikeDHS, NeuEvo, and STTS, demonstrating that evolutionary SNNs can achieve competitive accuracy on static and event-based vision tasks while reducing spike counts and energy use. The review highlights both the promise and current limitations—especially the high computational cost of evolution and the tendency to rely on convolution-centric operation spaces—calling for co-evolution of multiple design elements and broader search spaces. Overall, evolutionary SNNs offer a compelling path toward hardware-friendly, biologically plausible networks, with significant potential for future advances through more efficient search strategies and integrated optimization of architecture, dynamics, and learning.

Abstract

Spiking neural networks (SNNs) are gaining increasing attention as potential computationally efficient alternatives to traditional artificial neural networks(ANNs). However, the unique information propagation mechanisms and the complexity of SNN neuron models pose challenges for adopting traditional methods developed for ANNs to SNNs. These challenges include both weight learning and architecture design. While surrogate gradient learning has shown some success in addressing the former challenge, the latter remains relatively unexplored. Recently, a novel paradigm utilizing evolutionary computation methods has emerged to tackle these challenges. This approach has resulted in the development of a variety of energy-efficient and high-performance SNNs across a wide range of machine learning benchmarks. In this paper, we present a survey of these works and initiate discussions on potential challenges ahead.