METHOD FOR EXECUTION SPEEDUP OF NEURAL NETWORK FOR CORRELATION SIGNAL PROCESSING

    公开(公告)号:JPH08272759A

    公开(公告)日:1996-10-18

    申请号:JP9174596

    申请日:1996-03-22

    Abstract: PROBLEM TO BE SOLVED: To provide a method for increasing an execution speed of a neural network for a correlation signal. SOLUTION: By this method, an execution speed can be upped in most of a neural network to process input signals, e.g. a voice, radar, sonar, a video signal, etc., which temporally and slowly change, and furthermore, a specialized, expensive hardware or one that is hard to acquire is not needed. In the method, activation value is stored at one time and the activation value is requested to compare with value which is counted at the next time for a neuron (i) on at least one level of the network. When the activation value is equal, the neuron (i) does not act and when it is not, what difference of the activation value is multiplied by link weight Wij is transferred to connected neuron j.

Patent Agency Ranking