Time Delay Parameter Optimization Of Spiking Neural Network

Spike neural networks have neuron structures closer to biological neurons than traditional artificial neural networks.The transmitted signals between neurons contain more time information.Therefore,spiking neural networks can obtain more information and more computing power than traditional artificial neural networks.The time information in spiking neural network transmission signals can be traced back to the time delay characteristic of the conversion process of chemical substances between biological neurons.This kind of time delay has a special effect on the ability of the spiking neural network to process information.Based on the time delay characteristics,this paper is based on the spiking neural network model of the liquid state machine to study synaptic decay time???in the process of synaptic integration and the post-neuronal synaptic information transmission delay time???in the liquid layer.The main contents include:First,the influence of the internal state and performance on the spiking neural network was analyzed and studied,including the discharge complexity and the discharge synchronization of the neurons in the spiking neural network.The calculation results showed that with the increase of?,the complexity of neuron spiking discharge increased,and the expression ability of the characterization network was stronger.But the increase of?would make the interval of the spiking discharge time of the neural network smaller,reducing the synchronization,affecting the ability of the network to reduce noise.Secondly,for the reconstruction task of time series,the output Mean Squared Error?MSE?was used as a performance index to study and analyze the influence of the delay parameters?and?on the network computing capacity.From the four groups of time series ₁r,₂r,₃r,₄r,and their addition,multiplication,and squaring function reconstruction experiments,we knew that:1)With the increase of?,MSE would be significantly reduced to a low amplitude oscillation range;2)However,when increasing to a certain value,the MSE no longer decreased monotonically and lost the optimization value;3)The tendency of variation rule of MSE as a function of the increase of?was similar to case of increase of?Finally,by summarizing the performance of various time delay parameters,the basic steps for the optimization of time delay parameters are:1)Through the changing law of MSE,choose a suitable one in the low amplitude oscillation interval;2)Comprehensive discharge complexity,network synchronization performance,and the MSE of?variation rule determine the interval of?that satisfies these three indicators at the same time,and select a suitable value.