| 摘要 |
PURPOSE:To obtain an approximate solution in a short time without performing any calculation based on an approximate equation by using a neural net to learn a parameter and its corresponding solution as the teacher patterns. CONSTITUTION:A neural circuit is used as a model and at the same time a neural net model 11 consisting of an input layer, an output layer, and a middle layer is used. Thus a parameter and its corresponding solution are set as an input and an output respectively. Then a certain parameter obtained by a teacher pattern production mechanism 13 and the solution corresponding to the obtained parameter are used as the teacher patterns. The network weight is learnt so that the error caused between the output against the input of the teacher pattern and the output of the teacher pattern is minimized. In this case, several teacher patterns are applied and a network is learnt. Thus it is expected to obtain a solution approximate to the correct one to some extent as an output even to an unlearnt input. As a result, an approximate solution is obtained in a short time without performing any calculation based on an approximate equation. |
