Max-min propagation nets: learning by delta rule for the Chebyshev norm

Author

Estevez, Pablo A. ; Okabe, Yoichi

Author_Institution

Res. Center for Adv. Sci. & Technol., Tokyo Univ., Japan

Volume

1

fYear

1993

fDate

25-29 Oct. 1993

Firstpage

524

Abstract

This paper introduces max-min propagation nets, which are composed of multiple-input maximum (max) and minimum (min) operators, besides linear weighted sum units. A learning algorithm based on gradient descent is derived for these networks. Two different criteria of error measurement are tested: the Chebyshev norm and the least squares norm. Weight-update rules for both criteria are deduced and implemented together with acceleration techniques to speed up convergence. A comparison of results on the parity problem is presented.

Keywords

Chebyshev approximation; convergence of numerical methods; learning (artificial intelligence); minimax techniques; neural nets; Chebyshev norm; convergence; delta rule; error measurement; gradient descent; learning algorithm; least squares norm; linear weighted sum units; max-min propagation nets; maximum operators; minimum operators; parity problem; weight-update rules; Acceleration; Boolean functions; Chebyshev approximation; Convergence; Hardware; Least squares methods; Multi-layer neural network; Neural networks; Neurons; Testing;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Networks, 1993. IJCNN '93-Nagoya. Proceedings of 1993 International Joint Conference on

Print_ISBN

0-7803-1421-2

Type

conf

DOI

10.1109/IJCNN.1993.713968

Filename

713968