Surrogate Gradient Learning in Spiking Neural Networks

TL;DR

This paper discusses the challenges of training spiking neural networks and introduces surrogate gradient methods as an effective solution for enabling real-world applications.

Contribution

It provides a comprehensive overview of training challenges in spiking neural networks and introduces surrogate gradient methods as a novel, flexible approach to address these issues.

Findings

Surrogate gradient methods improve training efficiency.

They enable spiking neural networks to solve real-world problems.

The approach overcomes key challenges in synaptic plasticity and data-driven learning.

Abstract

Spiking neural networks are nature's versatile solution to fault-tolerant and energy efficient signal processing. To translate these benefits into hardware, a growing number of neuromorphic spiking neural network processors attempt to emulate biological neural networks. These developments have created an imminent need for methods and tools to enable such systems to solve real-world signal processing problems. Like conventional neural networks, spiking neural networks can be trained on real, domain specific data. However, their training requires overcoming a number of challenges linked to their binary and dynamical nature. This article elucidates step-by-step the problems typically encountered when training spiking neural networks, and guides the reader through the key concepts of synaptic plasticity and data-driven learning in the spiking setting. To that end, it gives an overview of existing approaches and provides an introduction to surrogate gradient methods, specifically, as a particularly flexible and efficient method to overcome the aforementioned challenges.