Systems Analysis

Nonlinear least squares method and block recurrent and iterative procedures in neural networks teaching

P. V. Saraev

Lipetsk State Technical University, Lipetsk

Abstract: The article is devoted to the development of the feedforward neural networks teaching algorithms based on nonlinear least squares method with pseudo-inversion. Application of Cline block pseudoinversion formula to teaching algorithms based on weights vector decomposition is considered. This approach allows to decrease computational costs by virtue of pseudo-inversion of matrices of small size.

Keywords: neural networks teaching, nonlinear least squares method, block pseudo-inversion, block recurrent and iterative procedures.

UDC: 519.7
BBK: 22.19