We present an Interactive Evolutionary Computation (IEC) system that applies user gaze information. Historically, IEC systems have encountered the problems caused by heavy user evaluation loads. To solve these problems, researchers have employed biologically derived user information, such as heartbeats or brainwaves, to reduce the evaluation load. However, the requirement for users to wear special devices to measure this information has limited the popularity of these systems. Therefore, we applied the user gaze information approach to solve these problems. Gaze information includes the user's potential preferences, which are derived from various processes. When user gaze information is applied in the evaluation of candidate solutions, IEC systems can obtain user evaluation information while users are viewing multiple candidate solutions. In this paper, we verify the effectiveness of the eye tracking IEC system using evaluation experiments with real users. In the experiment, we use a normal IEC system as a comparison method where users manually evaluate candidate solutions using a 10-stage evaluation process. The experimental results show that the eye tracking IEC method can generate solutions with results equivalent to those of the compared system.

No abstract

No abstract

Abstract:We describe the effectiveness of interactive evolutionary computation (IEC) with tournament-style evaluation, aiming to reduce the evaluation load of IEC users. In tournament-style evaluation, users evaluate candidate solutions using a paired comparison. We use three evaluation objects (music, animation, and images) and evaluate the performance of three methods. The first is a normal interactive genetic algorithm (NIGA), which is a conventional 10-stage evaluation. The second is a tournament-style evaluation with two levels (T2), which evaluates only the superiority or inferiority of two candidates. The third is a tournament-style evaluation with four levels (T4), which progressively evaluates the superiority or inferiority of two candidates. We inspect the effectiveness of tournamentstyle evaluation by numerical simulations using an evaluation agent that imitates human's Kansei and an evaluation experiment with real users. The simulation results and experimental results show that T2 and T4 more effective for reducing evaluation load of IEC users than NIGA.

No abstract