Synaptic weight update in deep spiking neural networks

Abstract

Biological Neurons show a very rich range of dynamic properties and working, whereas the neurons in Artifi cial Neural Networks though being a very crude approximation of these biological networks are nowhere near as versatile as the biological neurons. Besides this argument, there are many reasons like biologically implausible weight updating algorithm Back-propagation which refers to a notion of derivative/gradient of a neuron that is not biologically synonymous to neural networks. The main idea of this research is to use the power Dynamical System representation of a neuron for an e active synaptic weight update algorithm for developing a biologically plausible Spiking neural network algorithm. Proof of concept is shown empirically by showing that the proposed SNN architecture is able to learn image patterns reasonably well.