Spiking neural network with synaptic plasticity for recognition

Abstract

The spiking neural network referred to the third generation of neural network simulates the mechanisms of neurons and networks in brain. It has the distributed computational mechanism and robust information processing way like the nervous system. This paper describes that a spiking neural network with the synaptic plasticity recognizes the input scenes. The Digital spiking silicon neuron (DSSN), a mathematical structure-based qualitative model, is used to reproduce the various behaviors of neurons. We also designed the synapse model in our spiking neural network to precisely describe the dynamics of the transmitter release and the postsynaptic current generation. There are three layers in our network. The spiking neurons in layer 1 and 2 with special receptive fields perform the edge detection and orientation selection, respectively. The synaptic plasticity is realized in synaptic connections between spiking neurons in layer 2 and the output layer. The changing of connection is based on the Hebbian learning rule which supposes that the time difference of two spikes modifies the value of connection. We evaluated our spiking neural network with the task of image recognition. The spiking neurons in the output layer fire with the high frequency in response to their relevant input scenes. The simulation results show that our spiking neural network can successfully recognize the input scenes learned before. The recognition is robust against various distortions.