摘要 |
PURPOSE:To shorten learning time by dispersively executing the learning of a neural network at plural computers. CONSTITUTION:This neural network learning device is composed as a multilayered type of an input layer composed of plural nodes to a pattern, output layer composed of plural nodes to output the pattern, and intermediate layer composed of plural nodes. In order to divide the neural network, a computer 1 inputs network information 9 and an input/output pattern 8. Next, the computer 1 communicates computers 2-N connected to the own computer 1 and measures the performance of each computer and communicating time. Then, the computer 1 decides a dividing method according to measured data and the information of the neural network. The respective computers divide the neural network according to the decided dividing method and dispersively execute learning while communicating one and another.
|