Low-activity supervised convolutional spiking neural networks applied to speech commands recognition

TL;DR

This paper demonstrates that supervised spiking neural networks with leaky integrate-fire neurons can achieve speech command recognition accuracy close to traditional deep neural networks while maintaining very sparse activity, enhancing energy efficiency.

Contribution

It introduces a novel regularization term for sparse spiking activity and shows the effectiveness of leaky neuron models in speech recognition tasks.

Findings

Achieved 5.5% error rate on Google SC v1 dataset

Maintained below 5% spiking activity with regularization

Leaky neuron models outperform non-leaky models significantly

Abstract

Deep Neural Networks (DNNs) are the current state-of-the-art models in many speech related tasks. There is a growing interest, though, for more biologically realistic, hardware friendly and energy efficient models, named Spiking Neural Networks (SNNs). Recently, it has been shown that SNNs can be trained efficiently, in a supervised manner, using backpropagation with a surrogate gradient trick. In this work, we report speech command (SC) recognition experiments using supervised SNNs. We explored the Leaky-Integrate-Fire (LIF) neuron model for this task, and show that a model comprised of stacked dilated convolution spiking layers can reach an error rate very close to standard DNNs on the Google SC v1 dataset: 5.5%, while keeping a very sparse spiking activity, below 5%, thank to a new regularization term. We also show that modeling the leakage of the neuron membrane potential is useful, since the LIF model outperformed its non-leaky model counterpart significantly.