Entropy-based Reduction Operator for Heuristic Binary Optimization

In this paper, we propose a reduction operator addressing discrete optimization problems, more precisely, for optimization problems with binary represented decisions. This operator may be integrated within various metaheuristics in order to reduce the dimension of the target problem. This is, by performing an iterative supervision over the browsed admissible space throughout the search process. We thus use the information entropy concept to measure the current uncertainty towards decision variables current values provided by an iterative heuristic search algorithm. Hence, it allows to explore the decision space more efficiently. Furthermore, we present two frameworks as effective applications of this operator: a pre-process for a hybrid approach and an interactive heuristic approach. We compare it against established approaches from the literature.