Article at GECCO 2021
Article “On the Design and Anytime Performance of Indicator-based Branch and Bound for Multi-objective Combinatorial Optimization”, A.D. Jesus, L. Paquete, A. Liefooghe, B. Derel was accepted to GECCO 2021
Abstract:
In this article, we propose an indicator-based branch and bound (I-BB) approach for multi-objective combinatorial optimization that uses a best-first search strategy. In particular, assuming maximizing objectives, the next node to be processed is chosen with respect to the quality of its upper bound. This quality is given by a binary quality indicator, such as the binary hypervolume or the $\epsilon$-indicator, with respect to the archive of solutions maintained by the branch and bound algorithm. Although the I-BB will eventually identify the efficient set, we are particularly interested in analyzing its anytime behavior as a heuristic. Our experimental results, conducted on a multi-objective knapsack problem with 2, 3, 5, and 7 objectives, indicate that the I-BB can often outperform the naive depth-first and breadth-first search strategies, both in terms of runtime and anytime performance. The improvement is especially significant when the branching order for the decision variables is random, which suggests that the I-BB is particularly relevant when more favorable (problem-dependent) branching orders are not available.