| 摘要 |
PROBLEM TO BE SOLVED: To make it easy to structure a large-scale neural network. SOLUTION: The large-scale neural network is composed of blocks of function units. Interconnection relations between internal structures of the respective blocks and the respective blocks is described as block diagram data 21. Neural networks are generated from pieces of block diagram data 21 respectively and made to learn. An evaluating and sorting mechanism 26 evaluates the performance of each neural network having learnt to select superior ones. A chiasma and mutation mixing mechanism 27 handles respective pieces of block diagram data 21 corresponding to the selected neural networks as genes and handles descriptions of individual blocks included in the data 21 as genetic units to generate descendants from those superior block diagram data 21 by genetic algorithm. COPYRIGHT: (C)2004,JPO |
