Output details

A theoretical and empirical study on unbiased boundary-extended crossover for real-valued representation

<12> This collaboration with Seoul National University is published in Information Sciences, a premier journal with high impact factor.

Crossover operators for continuous spaces have an inherent bias which is detrimental for the search, and many modified crossovers exist that attempt heuristically to remove the bias (e.g., extended line crossover). This work, for the first time, proves rigorously the topological origin of the bias, and shows that traditional crossover operators on a topologically transformed space have provably no bias. These operators are shown to have superior performance on a broad test-set of problems.