Research And Application Of Dynamic And Self-Adaptive Model Based On Spiking Neural Network

As the third generation artificial neural network,Spiking neural network(SNN)is more biologically plausible.SNN has experienced an increasingly large amount of research attention.SNN consists of the spiking neurons,which are more biomimetic and communicate with each other using precise time of spike emission.Therefore,SNN has stronger biometric and more powerful computing capabilities.The coding mechanism,the spiking neuron and the topology of SNN are introduced.This is also the research direction that most scholars devoted to SNN are devoted to.Based on the analysis of SNN,this paper focuses on Evolving spiking neural network.This paper proposes a learning algorithm for Evolving spiking neural network and its application on pattern recognition,including as the followings.(1)This paper proposes a fast Precise-Spike and Weight-Comparison based learning approach for dynamic and self-adaptive model based on SNN.The proposed SNN consists of an encoding layer and an output layer.The encoding layer temporally encodes real-valued features into spike patterns,and the output layer dynamically grown neurons which perform classification.The encoding layer choses Gaussian receptive fields and square cosine population encoding schemes.Unlike the rank order based learning approach,this algorithm can dynamically add a new neuron or update the parameters of existing neurons according to the precise time of the incoming spikes and the similarities of the weights.(2)The proposed two-layer dynamically adaptive SNN is extended to three layers.The three layers' structure is encoding layer-hidden layer-output layer.The hidden layer is a dynamic layer,and the output layer is a fixed layer.The learning phase can be divided into two parts.The first part is parameter learning,including fine-tuning the weights of neurons in hidden layer and learning the weights of neurons in output layer,and the weights of neurons in output layer use STDP learning rules.The second part is structural learning dynamically.(3)The proposed dynamically adaptive models based on SNN are applied to the problem of pattern recognition.Benchmark datasets from the UCI machine learning repository are selected to test and verify the two models.The experimental results demonstrate that the first model has a significant advantage in terms of speed performance and provides competitive results in classification accuracy,and the second model also achieves ideal accuracy.The last application of the two models is face recognition.It not only enriches the application of dynamically adaptive model in pattern recognition,but also brings a new research direction for face recognition.