Abstract

Constraint satisfaction is the core of a large number of problems, notably scheduling. Because of their potential for containing the combinatorial explosion problem in constraint satisfaction, local search methods have received a lot of attention in the last few years. The problem with these methods is that they can be trapped in local minima. GENET is a connectionist approach to constraint satisfaction. It escapes local minima by means of a weight adjustment scheme, which has been demonstrated to be highly effective. The tunneling algorithm described in this paper is an extension of GENET for optimization. The main idea is to introduce modifications to the function which is to be optimized by the network (this function mirrors the objective function which is specified in the problem). We demonstrate the outstanding performance of this algorithm on constraint satisfaction problems, constraint satisfaction optimization problems, partial constraint satisfaction problems, radio frequency ...

Citations