Efficient multi-spike learning with tempotron-like LTP and PSD-like LTD

Abstract

Biological neurons use electrical pulses to transmit and process information in a significantly efficient way. To understand the mysteries of the underlying processing principles of the biological nervous systems, spiking neurons have been proposed to process information in a brain-like way. However, how could neurons learn spikes in an efficient way still remains challenging. In this study, we propose a simple and efficient multi-spike learning rule which could train neurons to associate input spike patterns with different output spike numbers. Our learning algorithm adopts a Tempotron-like LTP and a PSD-like LTD to adapt neuron’s efficacies. The results show that the proposed rule is faster than other benchmarks for the given task. A fast running time and simple implementation can largely benefit applied developments in neuromorphic systems. Additionally, we show that neurons with our proposed rule can elicit different output spike numbers in response to input spike patterns. Thus, single neurons are capable of performing the challenging task of multi-category classifications.