Multi-objective Evolutionary Top Rank Optimization with Pareto Ensemble

Kai Wu, Jing Liu

Abstract: The accuracy of the selected instances ranked near the top is a key issue for many information retrieval systems. Most existing approaches optimize the convex surrogate of the corresponding non-convex optimization problem, leading to the final solutions far from being completely and precisely consummated. In this paper, we establish a multi-objective top rank model for this non-convex optimization problem and propose a multi-objective evolutionary algorithm to solve this model. Furthermore, instead of using the widely used knee point-based method, we design a new ensemble method to determine the final solution based on the solutions in the obtained Pareto fronts. The performance of the proposed approach is evaluated on several binary classification datasets. Experimental results show that the proposed approach is highly competitive to the three state-of-the-art approaches.