Fuzzy based Modified SHL algorithm for Spiking Neural Networks

Spiking neural network is the 3 generation neural network. In this paper, we derive spiking neural network‘s topology and the fuzzy reasoning by restricting to the usage of biological components. Input encodes information in the timing of spike train. Fuzzy reasoning is used on biological components such as dynamic synapse, receptive field, inhibitory and excitatory neurons. The enrichment of the flow of information is done by dynamic synapse and the neuron selection by using receptive field. Modeling the dynamics of the limited synaptic resources makes neurons selective to particular spike frequencies. The receptive field behaves like fuzzy membership function which enables the individual neuron respond at certain spike train frequency. The network is supervised and learning occurs at the output layer of the network. Various issues arise while learning with supervised method takes place, namely convergence and continuous updating of weights after the goal is achieved. These issues are discussed in detail and are resolved. The modified SHL algorithm is used for learning. The classification problem of XOR is solved. The implementation is done on MATLAB.