Method of speeding the execution of neural networks for correlated signal processing

摘要

the method enables the speeding up of the implementation of a large class of neural networks for the treatment of signs of entry to slow time evolution, such as, for example, voice messages.radar, sonar, video, and does not require specialized hardware, expensive or difficult to call.the method requires to store for each neuron of the network, the value of activation at a certain moment and to compare with calculated instantly next.if the activation is equal, the neuron is not a business, otherwise it propagates the difference of activation in neurons which belongs, multiplied by the weights of interconnection.