抄録
This paper demonstrates that a recognition mechanism based on a biological one can be useful for recognizing “unknown” patterns, and also useful for self-learning of them. An essential point of our proposed mechanism is a dynamical recognition using chaotic dynamics of recurrent neural networks. Harnessing the complex dynamics, the networks can recognize the “known” patterns and their neighbors as the conventional recognition methods are possible. We present some simulation results illustrating that our networks are able to decide whether input patterns are “known” or “unknown” by observing temporal stability of output patterns. In addition, it is shown that recognition of “unknown” patterns makes it possible the networks to learn the new patterns automatically.
