Offline Design of Interactive Evolutionary Algorithms with Different Genetic Operators at Each Generation

Abstract

In interactive evolutionary computation (IEC), each solution is evaluated by human users. This means that only a small number of solutions can be evaluated during the execution of an IEC algorithm. In some application fields of IEC such as hearing aid design and music composition, only a single solution is evaluated at a time. Moreover, accurate and precise fitness evaluation is not easy for human users. Based on these discussions, we formulated an IEC model with the minimum requirements for human users’ fitness evaluation ability: They can evaluate only a single solution at a time, they can memorize only a single previous solution, and their evaluation result on the current solution is whether it is better than the previous solution or not. In this paper, we propose a meta-level approach to the design of interactive algorithms for our IEC model. The novelty of our approach is to use a different operator to generate each solution. An IEC algorithm is coded by a string of operators to generate solutions such as random creation, mutation and crossover. The string length is the same as the number of solutions to be generated. Each string (i.e., each IEC algorithm) is evaluated through simulations in our IEC model. A population of strings is evolved by a meta-level evolutionary algorithm. We perform experimental studies using a well-known test function as a surrogate of a human user. For a simple test function, we obtain (1+1)ES-style algorithms where the next solution is generated by mutation from the best solution among the examined ones. For a complicated test function, we obtain (μ+1)ES-style algorithms with crossover, mutation and random creation.