Stochastic Resonance Phenomenon Of Perceptual Neural Networks

Along with the improvement of living standards, the quality of life has become the new pursuit of the modern humans. But in this noisy society, sub-health and even diseases distress people due to many reasons, such as the mental pressures form all aspects person and personal accidents. We note that the problems in the nervous system, such as hearing-impaired, eyesight problem and poor tactile sensation, are prominent physical impairments. It is now known that noise exists in the neural system extensively, and play a positive role in nervous information processing, this is stochastic resonance. Since its introduction, the feasibility of stochastic resonance in improving the information processing of human becomes an interesting conceivability.In this paper, we use the measure of the average mutual information to explore the performance of information transmission of the FHN neural network and the hierarchical IF neural network. It is noted that the input sinusoidal signal and the internal noise components are interactional. First, it use the aperiodic signal as the input signal, and studies the transmission of information-carrying signals in sensory FHN neuron model by the measure of average mutual information. Second, we also take the measure of the average mutual information to explore the stochastic resonance in the hierarchical IF neural network by using the periodic signal as the input signal. Third, based on the mathematical characteristic of neuron model, we establish the nonlinear circuit of FHN neural network by using PSpice. The obtained results show that both FHN neural network and hierarchical IF neural network exhibit the stochastic resonance effect, which is also demonstrated in circuit experiments via PSpice software. It is proved that, as the internal noise intensity increases, the neurons have a better response to the input signal, wherein the average mutual information can reach a maximum value in a certain range of noise intensity. Therefore, the transmission efficiency of the neural network can be optimized at an optimal non-zero noise level. We argue that the present results are meaningful to the information-carrying signal transmission in sensory neurons.