Pattern classification based on a multi-spike learning algorithm in a photonic spiking neural network with VCSEL-SA

In this paper, we propose a pattern classification method based on the modified multi-spike Tempotron-like ReSuMe algorithm in a VCSEL-SA-based photonic spiking neuron network. Based on the multi-spike triggering mechanism, the proposed method can capture the global information to overcome the limitation of the traditional single-spike triggering algorithm, which can be used to effectively process more complex temporal information tasks, accompanied by good robustness to noise. The pattern classification task for the digits "1" to "4" demonstrates the superior performance of the proposed method in the information processing task. By adopting the bias current management strategy for the post-synaptic neuron, we can further improve the network's noise robustness. Moreover, this proposed method is validated in a pattern classification task in the Wisconsin Breast Cancer (WBC) dataset, and a classification accuracy of 95.6% can be achieved.