Modeling And Simulation Analysis Of Spiking Neural Networks Topology

The environment is filled with all kinds of electromagnetic field by the information technology. If the electronic system runs for long time, the electromagnetic field will cause component failure resulting in decreased system performance. It has taken the traditional electromagnetic protection means, such as grounding, shielding and filtering to minimize electromagnetic interference to electronic systems, but how to improve the electronic system reliability fundamentally is a thorny issue. Faced with the complex electromagnetic environment, organisms have demonstrated their characteristics, such as self-organizing, self-healing and self-adaptive, which are closely associated with the regulation of nervous system. The third generation artificial neural networks—spiking neural networks for electromagnetic protection are provided. The thesis focuses on the study of the relationship between the stability of output and the interference to the input layer of the spiking feedforward neural networks with STDP mechanism. Main works have been listed as followed:(1)The research on single neuron. By adding noise and reducing the number of presynaptic inputs to postsynaptic neuron in multiple presynaptic inputs—single postsynaptic output model, the impact on the encoding of postsynaptic neuron is analyzed. As a result, the output firing rate of the postsynaptic neuron changes little with the noise added to current and reducing the presynaptic inputs.(2)The research on robust of spiking neural network. Construct multi-layers feedforward spiking neural networks, conduct simulation experiment by introducing noise to input layer and damaging the nodes in input layer. Our works show that the feedforward spiking neural networks with STDP mechanism is robust to input noise and to a certain degree of damaged nodes.