Optimization of replica exchange method by SMBO

Abstract

The purpose of this research is to use Sequential Model-Based Global Optimization (SMBO) to adjust parameter values of replica exchange method sequentially. In the conventional replica exchange method, we need to determine the optimum number of replicas and the upper and lower limits of temperature. Therefore, in this study, we try to reduce the parameters setting load of the user and improve the search accuracy by updating the parameters to the optimum values by SMBO.