PATTERN RECOGNITION DEVICE AND ITS METHOD

PROBLEM TO BE SOLVED: To recognize relation between a newly presented pattern and an original stored pattern group by recovering a stored state without especially requiring timing supply from the external in the case of changing a presented pattern. SOLUTION: Plural stored network states 10a to 10d corresponding to respective stored patterns are stored in a mutual connection type neural network. When the network state 14 of the mutual connection type neural network has high similarity to any one of the states 10a to 10d and the stored network states of the similarity has low similarity to a presented network state corresponding to a presented pattern, the stored network state is excluded from the stored state of the neural network. On the other hand, the stored state of the neural network is gradually recovered to the original stored state with the lapse of time. Then relation between the presented pattern and the stored pattern is recognized based on the network state of the neural network.