Study On Stimulus Condition Information In Super-threshold Stochastic Resonance

In this paper,we study the phenomenon of super threshold stochastic resonance in the information transmission of threshold perception neural network with the stimulus specific information as the evaluation index.In the first,we explain and compare two evaluation methods which are the average mutual information and stimulus specific information.Specific information of stimulating is a decomposition form of the average mutual information in the signal space which played a similar role of "magnifying glass" and it can more reflect the stimulation signal encoding efficiency in signal transmission.Based onthe theory of the stimulus specific information,we use a new information decomposition form to analyze the mechanism of supra-threshold stochastic resonance.Using stimulus specific information,the stochastic resonance phenomenon is studied in the three-level neural network structure model as a function of noise intensity and structure caused by the network layer.The results show that the noise can improve the input-output threshold neuron network mutual information,and the average transmission performance can obtain the optimum maximum mutual information via tuning noise existing in the network.It is shown that the stimulus specific information is a new method to more effectively explain the observation of supra-threshold stochastic resonance mechanism.