Research And Application Of Spiking Neural Networks Learning Algorithm

In order to further study the ability of biological brain to process information and learning,researchers have proposed artificial neural networks to mimic the process of brain how to represent and process information,and the highest biomimetic is the spiking neural network,which uses spike-time coding to represent and transmit information.Compared with traditional neural networks,such as perceptions,spike neural networks are closer to biological brain neurons in in the aspect of mechanism of processing information.Many studies have shown that the spike neural network,whether in the ability of expressing information or computing power,are superior compared with the traditional neural network.As a result,many researchers at home and abroad has aroused widespread concern and attention about spike neural network.At present,the spike neural network have some preliminary research results in artificial intelligence areas and so on,but far from the level of commercial,relative to the second generation of neural networks such as deep learning,it is still less in the practical application.In addition,although the pulsed neural network has been proven to be the closest biological neural network,its learning mechanism of biological brain system is not clear,and the research is not mature about the learning process of neurons,so the study of learning methods is still a question worthy of study.In order to make full use of the features of spike neural network,such as highly bionic,strong the ability of representing information and computing power,this paper makes further research on the supervised learning algorithm.At present,there are some supervised learning algorithms,but learning efficiency or adaptability is still low.In order to improve the learning efficiency,accuracy and adaptability of the spike neural network,the some multi-layer network learning algorithm is optimized and extended by combining the classical algorithm such as ReSuMe algorithm and SpikeProp algorithm,and so on.and the algorithm is simulated and experimented.The work of this paper is as follows:1)Firstly,the advantages and disadvantages of some supervised learning algorithm,such as SpikeProp algorithm and ReSuMe algorithm,are analyzed.2)A multi-spike multi-layer spike neural network supervised learning algorithm isproposed.Combined with ReSuMe algorithm,the synaptic adjustment processof multi-spike multi-layer neural network supervised learning algorithm isdeduced in detail,and then the algorithm is optimized and innovated.3)On this basis,a spike neural network supervised learning algorithm based ondelay is proposed.the neural network learning process is no longer limited tothe adjustment of synaptic weights,and the existed algorithms is extended byincorporating the idea of delay.4)Finally,this algorithm is successfully applied to the XOR benchmark,Iris dataset classification and so on,showing a good effect.