Advancing Spatio-Temporal Processing in Spiking Neural Networks through Adaptation

TL;DR

This paper analyzes adaptive LIF neurons in spiking neural networks, revealing stability issues with traditional discretization and demonstrating that the Symplectic Euler method enhances performance on spatio-temporal tasks.

Contribution

It provides a thorough analysis of adaptive LIF neurons, identifies stability challenges, and proposes the Symplectic Euler method as an effective solution, improving performance on benchmark datasets.

Findings

Symplectic Euler method improves stability and performance

Adaptive LIF neurons excel at exploiting spatio-temporal input structures

Addressing discretization challenges enhances neural network robustness

Abstract

Implementations of spiking neural networks on neuromorphic hardware promise orders of magnitude less power consumption than their non-spiking counterparts. The standard neuron model for spike-based computation on such systems has long been the leaky integrate-and-fire (LIF) neuron. A computationally light augmentation of the LIF neuron model with an adaptation mechanism has recently been shown to exhibit superior performance on spatio-temporal processing tasks. The root of the superiority of these so-called adaptive LIF neurons however is not well understood. In this article, we thoroughly analyze the dynamical, computational, and learning properties of adaptive LIF neurons and networks thereof. Our investigation reveals significant challenges related to stability and parameterization when employing the conventional Euler-Forward discretization for this class of models. We report a rigorous theoretical and empirical demonstration that these challenges can be effectively addressed by adopting an alternative discretization approach - the Symplectic Euler method, allowing to improve over state-of-the-art performances on common event-based benchmark datasets. Our further analysis of the computational properties of networks of adaptive LIF neurons shows that they are particularly well suited to exploit the spatio-temporal structure of input sequences without any normalization techniques.