Anti-interference of a small-world spiking neural network against pulse noise.

Inspired by the nervous system working mechanism of a biological brain, brain-like intelligence has been a research frontier in the field of artificial intelligence. Under external stimulation, biological brains have self-adaptive advantages. Drawing from the advantage of biological brains, it is meaningful to investigate the anti-interference ability of brain-like models. In this study, we proposed a spiking neural network with small-world topology (SWSNN), where Izhikevich neuron models and synaptic plasticity models with excitatory and inhibitory synapses are introduced to represent nodes and edges of the network, respectively. The anti-interference ability of the SWSNN against pulse noise is investigated, and the anti-interference ability of SNNs with different topologies are compared. The simulation results indicate that: (i) our SWSNN has anti-interference ability against pulse noise, which is supported from different perspectives based on two indices. Furthermore, the chain reaction of firing rates, synaptic weights and topological characteristics forms neural information processing in the SWSNN under pulse noise. In addition, the synaptic weights are significantly relevant to the anti-interference ability, which implies that an intrinsic factor of the anti-interference ability is the dynamic regulation of synaptic plasticity. (ii) Our SWSNN outperforms the random and regular SNNs that are not complex networks in terms of anti-interference performance. For complex network, the anti-interference performance of SWSNN is superior to that of scale-free SNN, and the anti-interference superiority of the SWSNN is more obvious with the increase of amplitude of pulse noise. The topological characteristics of the network are further discussed, and the results imply that the topology is a factor that affects the anti-interference performance of the SWSNN. [ABSTRACT FROM AUTHOR]