Research On Information Transmission Of Neural Random Pool Network

This thesis employs the measures of the mean mutual information and the stimulus-specific information to explore the performance of information transmission of the stochastic pooling networks composed of saturating synaptic neurons or the Integrate-and-Fire neurons.It is noted that the stimuli are the deterministic aperiodic or the speech signals,while the internal noise is with Gamma or Gaussian distributions.First,in saturating synaptic neural networks,it applies the aperiodic signal as the input signal,selects Gamma noise to simulate the internal noise in nerve cells.The results of the measure of the mean mutual information and the stimulus-specific information demonstrate that the noise-enhanced effect of information transmission appears in the heterogeneous stochastic pooling networks with multi-synaptic excitatory and inhibitory pathways.Second,we also take the measure of the mean mutual information and the stimulus-specific information to explore the stochastic resonance effect in the Integrate-and-Fire neural stochastic pooling networks by transmitting the speech signal.The main purpose is to analyze the effect of information transmission caused by noise intensity and the number of the Integrate-and-Fire neurons.Third,according to the actual data experiment,it proves that,as the internal noise intensity increases,the neurons have a better response to the input signal,wherein the mean mutual information can reach a maximum value in a certain range of noise intensity.It is also shown that the stimulus-specific information can measure the coding efficiency in each component of the input signal via the internal noise enhancement clearly.We argue that the present results are meaningful to the information-carrying signal transmission in the neurons in future particularly.