An Online Supervised Learning Algorithm Based On Local Variable Driven STDP For Spiking Neural Networks

Spiking neural network is more in line with the requirements of the biological nervous system than traditional artificial neural network.It is suitable for the research and analysis of brain neural signals,and have powerful computing capabilities.Therefore,the related research and application of spiking neural network has received more and more attention.But there is a core question in neuroscience,that is,how does the brain achieve real and effective learning? In response to this problem,the synaptic plasticity mechanism is generally accepted by researchers.In recent years,neurophysiological studies have shown that changes of synaptic weights are closely related to the precise timing of neuron firing spikes.This phenomenon can be called Spike Timing Dependent Plasticity(STDP).However,for the STDP learning mechanism,it is very difficult to calculate the changes of a single synapse according to the general framework of synaptic plasticity,which requires a complete measurement and storage of its group activities.In addition,most of the data we obtain in the real world is real-time data,and online learning methods better learn and process these data.Therefore,an important research trend of spiking neural networks is to construct online supervised learning algorithms with high efficiency and wide applicability.This thesis introduces the synaptic plasticity mechanism,STDP learning rule and pair-based STDP learning rule,providing theoretical support for the successful derivation of the algorithm formula,and also proposing a theoretical foundation for the online supervised learning algorithm based on local variable driven STDP for spiking neural networks.The main tasks are as follows:(1)This thesis uses the spike response model,and propose an online supervised learning algorithm of spiking neuron synapse weights based on local variable STDP according to the Widrow-Hoff and STDP learning rules.The proposed algorithm is experimentally analyzed through a series of spike sequence learning tasks.And different parameter settings are selected to analyze their impact on learning performance.At the same time,the proposed algorithm is compared with online PSD,ReSuMe and STIPLR learning algorithms.The experimental results show that the online supervised learning algorithm based on the local variable STDP for spiking neurons can more effectively adjust the synaptic weights and improve the learning performance.(2)This thesis proposes an online supervised learning algorithm based on local variable driven STDP for spiking neural network,and introduces the spike sequence coding method of EEG signals in detail.At the same time,the proposed learning algorithm is applied to the epilepsy classification experiment based on EEG data to verify the effectiveness of the algorithm,and is compared with the classification results of some traditional feature extraction methods.Experimental results show that the proposed learning algorithm has good classification accuracy.