PURPOSE: To learn arbitrary association to quickly recognize the pattern by providing plural layers and providing them with plural neurons which act on layer input signals in parallel.
CONSTITUTION: A neural system simulating system 10 is provided with plural layers 12, and output signals of plural layers 12 out of these layers 12 become an input signal of another layer 12. Each layer 12 is provided with plural neurons which act on the input signal in parallel, and neurons are arranged in respective groups and have different functions. Respective neurons in one group act on the input signal in parallel and execute the same function to extract a group function from the area of an image to be processed. Each neuron induces the output signal, which indicates the relative weight of the applied input signal, from the input signal based on a continuous differential transfer function to each function. Thus, the neural system simulating network system which learns arbitrary association to quickly recognize the pattern is obtained.
Next Patent: DOCUMENT GENERATION DEVICE