An experimental investigation of self-adaptation in evolutionary programming

Abstract

Evolutionary programming (EP) has been widely used in numerical optimization in recent years. One of EP's key features is its self-adaptation scheme. In EP, mutation is typically the only operator used to generate new offspring. The mutation is often implemented by adding a random number from a certain distribution (e.g., Gaussian in the case of classical EP) to the parent. An important parameter of the Gaussian distribution is its standard deviation (or equivalently the variance). In the widely used self-adaptation scheme of EP, this parameter is evolved, rather than manually fixed, along with the objective variables. This paper investigates empirically how well the self-adaptation scheme works on a set of benchmark functions. Some anomalies have been observed in the empirical studies, which demonstrate that the self-adaptation scheme may not work as well as hoped for some functions. An experimental evaluation of an existing simple fix to the problem is also carried out in this paper.