Effect of Background Noise on Neuronal Spiking Reliability in Temporal coding-based Spiking Neural Network

The neuronal spike timing was usually supposed to be not robust enough to code neuron information because the background noise would influence neuronal precise spiking timing. In this paper, we verify the robustness of rank order coding, which is a type of temporal coding, under balanced background noise. We utilize the rank-order coding-based Spiking Neural Network (SNN) to analysis the effect of background noise, SNN is trained to recognize images by Spike-prop algorithm and tested under noise of different strength, the robustness is represented by recognition accuracy of images. The result shows that even if the noise would reduce the precision of neuronal spike timing, the SNN is not sensitive to the noise, it could conduct information correctly in the existence of strong background noise, meanwhile, adding background noise in training enable SNN to be more robust. We also reveal that weak synaptic coupling among neurons enables the SNN to be more robust through decreasing the fluctuation of neuronal spike timing.