NEURAL CIRCUITS FOR SOLVING NONLINEAR PROGRAMMING PROBLEMS

Abstract

This paper is concerned with neural networks which have the ability to solve linear and nonlinear constrained optimization tasks. After a short overview of such neural networks, we introduce useful extensions which make them capable of solving more general programming tasks, namely handling equality constraints more efficiently than in the known obvious ways, and obtaining global optimum with probability close to 1. We also refer to stability analysis worked out from both the circuit theory and optimization theory point of view. The simple simulation examples, one of which is presented in the paper, show that the extended networks are stable and converge to right solutions.