Towards learning principles of the brain and spiking neural networks

Anonymous

Keywords: spiking neural networks, Spike-time dependent plasticity, network simulations

Abstract: The brain, the only system with general intelligence, is a network of spiking neurons (i.e., spiking neural networks, SNNs), and several neuromorphic chips have been developed to implement SNNs to build power-efficient learning systems. Naturally, both neuroscience and machine learning (ML) scientists are attracted to SNNs’ operating principles. Based on biologically plausible network simulations, we propose that spatially nonspecific top-down inputs, projected into lower-order areas from high-order areas, can enhance the brain’s learning process. Our study raises the possibility that training SNNs need novel mechanisms that do not exist in conventional artificial neural networks (ANNs) including deep neural networks (DNNs).