Noise Perturbation Analysis Of Spiking Neural Networks

Being affected by many factors, the input signal of a neural network may produce some noise. Spiking neural network is the personification of strong performance of new type of network, generating the spikes through simulation between the dendrites and the axons processing all kinds of information. Due to noise interference, the ability of information processing of spiking neural network will be affected to a certain extent, such as the input signal has a deviation, the delay time is too long or too short, the threshold value is too large or too small, the value is too large or too small and so on. Spiking neural network is particularly strong for the sensitivity of the input signal, the input signal is disturbed, the processing power of the spiking neural network is vulnerable to the impact of disturbance. In this article, we study random perturbations within the range of values of the input signal to the spiking neural network is used to study the robustness of the classification ability. In this thesis we study the robustness of the classifying ability of a spiking neural network by adding random perturbations within the range of values of the input signal. A series of numerical experiments show that to add the random perturbations to the input signal substantially does not affect the classification ability of the network and the correct recognition rate difference between the network with disturbance and it without disturbance is just less than4%. But a more interesting result is that the recognition rate of the disturbed network is improved. This is just another proof which illustrates the strong personification of the spiking neural network of the abstract.