| 摘要 |
An application of chaos motion, which has been found recently in the field of physics, to a neural network. The chaos motion has both randomness and regularity. Using this feature ingeniously, an element for neural networks is made by coupling a plurality of chaos vibrators to each other. This new neural network can be realized only by replacing the elements in a conventional existing neural network with the chaos elements. The whole know-how relative to using conventional neural networks, e.g., the manner of coupling the elements, can be applied to the new neural network. Because of the use of the chaos vibrators, in spite of using no stochastic process such as the generation of a random number at all, the neural network can serve like a Boltzmann machine. Thus, pattern recognition and optimization problems where a great number of combinations are dealt with, and so on at which conventional computers are weak can be solved rapidly by the neural network operating as a highly parallel distributed system. |
