CIRM-SNN: Certainty Interval Reset Mechanism Spiking Neuron for Enabling High Accuracy Spiking Neural Network

Spiking neural network (SNN) based on sparse trigger and event-driven information processing has the advantages of ultra-low power consumption hardware friendliness. As a new generation networks, SNN is widely concerned. At present, most effective way to realize deep through artificial (ANN) conversion. Compared with original ANN, converted suffers from performance loss. This paper adjusts spike firing rate spiking neurons minimize loss in conversion process. We map ANN weights corresponding after continuous normalization, which ensures that neuron normal range. propose certainty interval reset mechanism (CIRM), effectively reduces membrane potential avoids problem neuronal over-activation. In experiment, we added modulation factor CIRM further adjust neurons. The accuracy CIFAR-10 1.026% higher than ANN. algorithm not only achieves lossless but also energy consumption. Our improves (VGG-15) CIFAR-100 decreases delay. work this great significance for developing high-precision depth SNN.