COTLEA: COpula-based Transfer Learning Evolutionary Algorithm for Multi-Objective Optimization Problems

A major assumption often made during the resolution of NP-Hard Multi-Objective optimization Problems is that problem instances are different from each other. Then, the same process is repeated integrally to solve these instances without benefiting from prior resolutions of similar problem instances. Consequently, resolution systems lose a lot of time to solve problems that were already partially or integrally treated beforehand. In this paper, we assume that previous Pareto Front solutions can guide new problem instance resolutions to accelerate the convergence to the optimal solutions. According to this assumption, our objective is to propose a framework that first learns a feature model on the previous solutions and second integrates this model in the resolution process. To this end, we introduce a new heuristic, called COTLEA, for COpula-based Transfer Learning Evolutionary Algorithm. COTLEA is a simple and fast transfer learning framework that incorporates a Copula model to learn the feature representation of a source problem and uses an evolutionary algorithm to make the transfer of knowledge. The empirical study of our proposal shows significant performance in terms of convergence speed compared to typical evolutionary approaches.