A description of a new algorithm for the automatic neural network construction technique that involves positive features of the evolutionary and constructive approaches is presented. The main idea of the algorithm is based on the following: at each step, a neural network is increased by adding blocks of new neurons having a maximal influence on the error reduction. The search for the blocks is implemented with genetic algorithms. The process of constructing a neural network ends up when the network's error achieves a prescribed value. The efficiency of this algorithm is validated with the use of benchmark problems.

• PostScript (in Russian) ( 2.1MB)
• PDF (in Russian) (1.04MB)



• PostScript. zip (in Russian) (619йB)
• PDF. zip (in Russian) (412йB)