A new recursive least squares-based learning algorithm for spiking neurons

摘要: Spiking neural networks (SNNs) are regarded as effective models for processing spatio-temporal information. However, their inherent complexity of temporal coding makes it an arduous task to put forward supervised learning algorithm, which still puzzles researchers in this area. In paper, we propose a Recursive Least Squares-Based Learning Rule (RLSBLR) SNN generate the desired spike train. During process our method, weight update is driven by cost function defined difference between membrane potential and firing threshold. The amount modification depends not only on impact current error function, but also previous functions evaluated weights. order improve performance, integrate modified synaptic delay proposed RLSBLR. We conduct experiments different settings, such spiking lengths, number inputs, rates, noises parameters, thoroughly investigate performance algorithm. RLSBLR compared with competitive algorithms Perceptron-Based Neuron (PBSNLR) Remote Supervised Method (ReSuMe). Experimental results demonstrate that can achieve higher accuracy, efficiency better robustness against types noise. addition, apply open source database TIDIGITS, show algorithm has good practical application performance.