Section: New Results
Neutrality in Multi-objective Local Search
Participants: Aymeric Blot, Clarisse Dhaenens, Laetitia Jourdan
External Participants: Hernan Aguirre, Kiyoshi Tanaka - Shinshu University, Japan
In multi-objective combinatorial optimization, the dominance-based local search algorithms are faced to sets of non-comparable solutions. In the absence of preferences, these solutions are equally good from the Pareto dominance perspective and can be considered neutral in term of quality, similar to the solutions who shares the same fitness value in mono-objective optimization. We propose two ideas to use the neutrality to improve the current local search algorithms. First, we analyze the distribution of neighbors for both small fully enumerable instances and hard large instances, to understand the distribution of neutral neighbors according to the rank of the solutions. Then, we compare the results of the proposed algorithms with the standard ones according to different indicators.