Effective calculations on neuromorphic hardware based on spiking neural network approaches

Abstract

The nowadays’ availability of neural networks designed on power-efficient neuromorphic computing architectures gives rise to the question of applying spiking neural networks to practical machine learning tasks. A spiking network can be used in the classification task after mapping synaptic weights from the trained formal neural network to the spiking one of same topology. We show the applicability of this approach to practical tasks and investigate the influence of spiking neural network parameters on the classification accuracy. Obtained results demonstrate that the mapping with further tuning of spiking neuron network parameters may improve the classification accuracy.