New arXiv guide: tuning LIF and RAF spiking neurons for NeuroAI

Spiking neural networks (SNNs) promise energy-efficient, event-driven AI, but getting their neuronal dynamics right has often been black magic. A new arXiv paper from researchers including William Gebhardt and Alexander G. Ororbia delivers a systematic tuning guide. The team focuses on the two most foundational spiking neuron models: leaky integrate-and-fire (LIF) and resonate-and-fire (RAF). They walk through the underlying mathematical equations, then show exactly how hyperparameter choices—such as membrane time constants, firing thresholds, and reset mechanisms—shift network behavior from stable to chaotic. The practical angle is strong: instead of abstract theory, readers get concrete guidance on how each parameter changes spike timing and rate.

The guide also moves beyond individual neurons to network-level design decisions that are often glossed over. It tackles input encoding strategies, like rate versus temporal coding, and explains how to construct neural assemblies—groups of neurons that fire together to represent information. The authors further cover the setup of excitatory-inhibitory (E-I) populations, a critical balance that influences oscillatory dynamics and signal propagation in biological and artificial circuits. This expanded v2, updated from June 2025 to August 2026, adds depth for researchers bridging computational neuroscience and NeuroAI. For anyone building SNN-based systems, this paper is a rare practical reference—one that turns trial-and-error tuning into a principled engineering process.