A Hybrid Learning Algorithm Fusing STDP with GA based Explicit Delay Learning for Spiking Neurons

Abstract

This paper presents a hybrid learning algorithm for spiking neural networks (SNNs), referred to as an evolvable spiking neural network (ESNN) paradigm. The algorithm integrates a supervised and unsupervised learning approach. The unsupervised approach exploits a spike timing dependent plasticity (STDP) mechanism with explicit delay learning for multiple connections between neurons. Supervision of the synaptic delays and the excitatory/inhibitory connections is governed by a genetic algorithm (GA), while the STDP rule is free to operate in its normal unsupervised manner. A spike train encoding/decoding scheme is developed for the algorithm. The approach is validated by application to the Iris classification problem