Fitness Noise in Interactive Evolutionary Computation and the Convergence Robustness

Noise is one of the important factors that influence the performance of evolutionary computation (EC). Many studies on noise were reported in traditional EC, but less in IEC. The reported work on noise in EC is reviewed firstly. Then the convergence robustness against fitness noise in IEC is studied secondly. Mapping among spaces, dominating relationship and convergence in IEC are discussed, which establish bases for two theorems -strong condition theorem and weak condition theorem. These two theorems imply that the fitness noise caused by a rational user will not prevent algorithm from converging to the optima. As the successive issue, the following conclusions are put forward: the effective fitness scaling is a case of the weak condition; the user preference is the true fitness in IEC. And, the narrow definition of fitness noise in IEC is also given. The experiments and the results validate the theorems. The results establish necessary foundation for future research

1. Introduction

Interactive evolutionary computation (IEC) is simply a kind of evolutionary computation (EC), in which the individuals' fitness is provided by a user's subjective evaluation [1]. In basic EC applications, the fitness of a solution usually takes a value that is certain and unchanging [2]. However, given the noisy nature of many real-world optimization problems, it is often an idealization [3]. How can one decide the existence of noise? If different results for repeated measurements of the same operating point are detected in unchanged environment, then there is noise [4]. However, the repeated same results don't mean there is no noise.