Random Spikes to Enhance Learning in Spiking Neural Networks

摘要

In spiking neural networks (SNNs), unlike traditional artificial neural networks, signals are propagated via a 'pulse code' instead of a 'rate code'. This results in incorporating the time dimension into the network and thus theoretically ensures a higher computational power. The different principle of operation makes learning in SNNs complicated, hi this paper, two ideas have been proposed to assure spiking activity propagation - random spikes and parallel input layers. The proposed ideas have been illustrated with experimental results.