R-Tempotron: A robust tempotron learning rule for spike timing-based decisions

The brain-inspired spiking neurons are more biologically plausible and computationally powerful than traditional rate-based artificial neurons. Most of the existing spiking neuron based applications assume noise-free condition for learning and testing. However, noise widely exists in the network of spiking neurons and the neural response can be significantly disturbed by different types of noise. Therefore, how to improve the anti-noise capability of spiking neurons remains an open question. In this paper, by analyzing the ways of noise disturbing the neural response, we put forward a robust Tempotron (R-Tempotron) learning rule for spike-timing based decisions. Simulation results demonstrate that the anti-noise capability of R-Tempotron is better than that of Tempotron, and the R-Tempotron can make a correct decision even under highly noisy conditions.