Interactions among temporal mechanisms in evolutionary spiking networks

Characterize the interactions among axonal conduction delays, synaptic time constants, and spike after-potential (bursting) parameters in feedforward spiking neural networks trained via evolutionary algorithms, and determine their computational advantages within an evolutionary context.

Background

The paper motivates studying temporal adaptation in spiking neural networks by noting that most prior work focuses on adapting only weights and biases, or at most a single temporal parameter, despite extensive biological evidence for plasticity in conduction delays, time constants, and bursting behaviors. The authors highlight that evolution has produced diverse neuron types whose morphology affects temporal processing, suggesting potential computational benefits when multiple temporal parameters are adapted in models.

Within this context, they explicitly state that a comprehensive understanding of how delays, time constants, and bursting interact, and what computational advantages these interactions confer under evolutionary training, remains unclear. This open problem frames the study’s broader aim to explore co-adaptation of multiple temporal mechanisms and informs implications for neuromorphic hardware design.