Interactive Evolutionary Computation with Evaluation Characteristics of Multi-IEC Users

We (1) propose a method for accelerating the convergence of interactive evolutionary computation (IEC) by incorporating multiple evaluation models of previous IEC users, (2) evaluate the method´s performance according to the similarity metric of users´ evaluation characteristics, and (3) investigate its practical usefulness by measuring users´ evaluation characteristics for real-world applications on the metric. Although conventional IEC with a function learning the current IEC user´s evaluation characteristics cannot use the evaluation characteristics until the model is learned, the proposed IEC uses models learned from previous users until the current user´s behavior is learned. The model from a previous IEC user whose evaluation values are most similar to those of the current IEC user is selected and used instead of the current IEC user´s model till the current user´s model is leaned. The viability of this method is evaluated on similarity distance of evaluation characteristics with simulation, and the simulation results are compared with the real IEC user´s evaluation characteristics for four different types of real applications. Through this evaluation, we obtain a rating method for predicting the effectiveness of the proposed acceleration method for different types of IEC applications.