Ulster University Logo

Ulster Institutional Repository

SWAT: A Spiking Neural Network Training Algorithm for Classification Problems

Biomedical Sciences Research Institute Computer Science Research Institute Environmental Sciences Research Institute Nanotechnology & Advanced Materials Research Institute

Wade, John, McDaid, Liam, Santos, Jose and Sayers, Heather (2010) SWAT: A Spiking Neural Network Training Algorithm for Classification Problems. IEEE Transactions on Neural Networks, 21 (11). pp. 1817-1830. [Journal article]

[img]
Preview
PDF - Published Version
952Kb

URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=5585775&tag=1

DOI: 10.1109/TNN.2010.2074212

Abstract

This paper presents a synaptic weight associationtraining (SWAT) algorithm for spiking neural networks (SNNs).SWAT merges the Bienenstock–Cooper–Munro (BCM) learningrule with spike timing dependent plasticity (STDP). TheSTDP/BCM rule yields a unimodal weight distribution wherethe height of the plasticity window associated with STDP ismodulated causing stability after a period of training. The SNNuses a single training neuron in the training phase where dataassociated with all classes is passed to this neuron. The rulethen maps weights to the classifying output neurons to reflectsimilarities in the data across the classes. The SNN also includesboth excitatory and inhibitory facilitating synapses which createa frequency routing capability allowing the information presentedto the network to be routed to different hidden layer neurons.A variable neuron threshold level simulates the refractory period.SWAT is initially benchmarked against the nonlinearly separableIris and Wisconsin Breast Cancer datasets. Results presentedshow that the proposed training algorithm exhibits a convergenceaccuracy of 95.5% and 96.2% for the Iris and Wisconsin trainingsets, respectively, and 95.3% and 96.7% for the testing sets,noise experiments show that SWAT has a good generalizationcapability. SWAT is also benchmarked using an isolated digitautomatic speech recognition (ASR) system where a subset ofthe TI46 speech corpus is used. Results show that with SWAT asthe classifier, the ASR system provides an accuracy of 98.875%for training and 95.25% for testing.

Item Type:Journal article
Keywords:Index Terms—Automatic speech recognition, Bienenstock– Cooper–Munro, dynamic synapses, spike timing dependent plasticity, spiking neural networks.
Faculties and Schools:Faculty of Computing & Engineering
Faculty of Computing & Engineering > School of Computing and Intelligent Systems
Research Institutes and Groups:Computer Science Research Institute
Computer Science Research Institute > Intelligent Systems Research Centre
ID Code:16364
Deposited By:Dr John Wade
Deposited On:11 Nov 2010 15:22
Last Modified:11 Nov 2010 15:22

Repository Staff Only: item control page