A Biologically Inspired Supervised Learning Rule for Audio Classification with Spiking Neural Networks
Date
2021-06-15
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Audio classification has many practical applications such as noise pollution monitoring, wildlife monitoring, audio surveillance, speech recognition, and more. For many of these applications, deploying classifiers on low powered devices for persistent monitoring is desirable. Artificial neural networks (ANN) have achieved significant success for audio classification tasks. However, it may not always be feasible to deploy current state-of-the-art ANNs to embedded devices due to their memory footprint and power consumption. Biologically inspired neural networks, also known as spiking neural networks (SNN), have been shown to significantly reduce power consumption during inference when compared with equivalent ANNs. They have also been theoretically proven to be more computationally powerful per unit than ANNs. These two properties make SNNs an attractive solution for machine learning tasks on low powered embedded devices, such as at the edge in an Internet of Things (IoT) sensor network. However, SNNs tend to lag behind in performance when compared to ANNs. This is partially because training SNNs is difficult since the standard backpropagation algorithm is not directly applicable due to the non-differentiable spiking nature of SNNs. Encoding data into spike trains compatible with SNNs is also an unresolved question when applying SNNs. This work compares different spike encoding schemes for audio data, and a learning algorithm for multilayer SNNs inspired by biologically plausible learning rules is developed. The proposed learning rule is then successfully applied to simple pattern recognition and audio classification tasks.
Description
Keywords
Spiking Neural Network, Classification, Machine Learning
Citation
Peterson, D. G. (2021). A Biologically Inspired Supervised Learning Rule for Audio Classification with Spiking Neural Networks (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.