Proposal of Model-Based Interactive Evolutionary Computation

Abstract

We previously confirmed through experiments that users of an interactive evolutionary computation system become more aware of their tastes as they frequently encounter their preferred solutions during the system’s execution. This fact suggested that for efficient solution search, the limited number of evaluations available to the user should be appropriately allocated between a stage in which the user becomes aware of their tastes and a stage in which the user intensively searches the area in which the preferred solutions exist. In this paper, we therefore propose an interactive evolutionary computation with such a two-stage procedure. This procedure is similar to model-based evolutionary computation, which extracts important variables from a set of selected good solution candidates and uses them to generate new solution candidates, and we call the proposed interactive evolutionary computation model-based interactive evolutionary computation. Furthermore, we develop a model-based interactive evolutionary computation system that performs simple image processing and use this system to evaluate the effectiveness of model-based interactive evolutionary computation.