Dynamic Mutation Based Pareto Optimization for Subset Selection

摘要

Subset selection that selects the best k variables from n variables is a fundamental problem in many areas. Pareto optimization for subset selection (called POSS) is a recently proposed approach for subset selection based on Pareto optimization and has shown good approximation performances. In the reproduction of POSS, it uses a fixed mutation rate, which may make POSS get trapped in local optimum. In this paper, we propose a new version of POSS by using a dynamic mutation rate, briefly called DM-POSS. We prove that DM-POSS can achieve the best known approximation guarantee for the application of sparse regression in polynomial time and show that DM-POSS can also empirically perform well.