Differential evolution and non-separability: using selective pressure to focus search

Recent results show that the Differential Evolution algorithm has significant difficulty on functions that are not linearly separable. On such functions, the algorithm must rely primarily on its differential mutation procedure which, unlike its recombination strategy, is rotationally invariant. We conjecture that this mutation strategy lacks sufficient selective pressure when appointing parent and donor vectors to have satisfactory exploitative power on non-separable functions. We find that imposing pressure in the form of rankbased differential mutation results in a significant improvement of exploitation on rotated benchmarks. Categories and Subject Descriptors I.2.8 [Artificial Intelligence]: Problem Solving, Control

• Type: Conference
• Year: 2007
• Where: GECCO
• Authors: Andrew M. Sutton, Monte Lunacek, L. Darrell Whitley
• Keywords: Differential Evolution Algorithm, Differential Mutation, Differential Mutation Procedure, GECCO 2007