Original scientific paper

Convex Optimization in Training of CMAC Neural Networks

Mato Baotić
Ivan Petrović
Nedjeljko Perić

Abstract

Simplicity of structure and learning algorithm play an important role in the real-time application of neural networks. The Cerebellar Model Articulation Controller (CMAC) neural network, with associative memory type of organization and Hebbian learning rule, satisfies these two conditions. But, Hebbian rule gives poor performance during off-line identification, which is used as a preparation phase for on-line implementation. In this paper we show that optimal CMAC network parameters can be found via convex optimization technique. For standard l2 approximation this is equivalent to the solution of Quadratic Program (QP), while for l1 or l‡ approximation solving Linear Program (LP) suffices. In both cases physical constraints on parameter values can be included in an easy and straightforward way.

Keywords

CMAC neural network; identification; convex optimization; quadratic program; linear program

Hrčak ID:

6617

URI

https://hrcak.srce.hr/6617

Publication date:

21.12.2001.

Article data in other languages: croatian

Visits: 1.757 *