Research On Spatio-temporal Brain Data Classification With Recurrent Spiking Neural Networks

Spiking neural network(SNN)is the third generation neural network which simulating the information strategy of human brain and brain system.The coding method of information in SNN uses precision timing coding method to convert the objective world information into spike sequence for research rather than using traditional frequency-based coding method,which is more conform to the information processing mechanism of the neural system.The study shows that the SNN is more accurate than the traditional neural network in simulating the human brain's neural system and is very suitable for the study of the problem of cranial process.However,it is very complicated to construct appropriate processing model because of the complexity of the related coding method.In this paper,we designs a new supervised learning algorithm and a classification model which can be used to classify spatio-temporal brain data(STBD)by referring to the model and learning method of feed forward spiking neural network structure.Firstly,We designs a new supervised learning algorithm which has good performance on recurrent neural network,and adjusting the synapse weight value by using this algorithm to different neuron in different layer.The experimental results shows that this algorithm has higher learning precision and better ability,and it is very effective in dealing with complex sptio-temporal.Secondly,according to the above method,a new supervised learning algorithm of recurrent spiking neural network with good applicability is proposed.The algorithm is applied to the neuron synapses corresponding to different layers of the network,and the synaptic weights are adjusted by error back propagation algorithm.After that we test the performance of this algorithm by applying to the spike sequence learning task.The experimental results shows that the algorithm has good adaptability,and it is very effective in dealing with complex time-spatio learning problems.Finally,we applied this classification model into Electroencephalogram(EEG)data and functional magnetic resonance imaging(fMRI)data to test the performance of proposed algorithm.In the experiment,these two different types data are converted into spike sequence by using different coding method and then we uses the obtained spike sequences as the input signal of this recurrent spike neural network.The experiment shows that this model has good performance on complex spatio-temporal pattern classification problem and has certain value in practical application.