"Optimal" mutation rates for genetic search

Using a set of model landscapes we examine how different mutation rates affect different search metrics. We show that very universal heuristics, such as 1/N and the error threshold, can generally be improved upon if one has some qualitative information about the landscape. In particular, we show in the case of multiple optima (signals) how mutation affects which signal dominates and how passing between the dominance of one to another depends on the relative height and size of the peaks and their relative positions in the configuration space. Categories and Subject Descriptors I.2.m [Computing Methodologies]: Artificial Intelligence-Miscellaneous General Terms Performance Keywords Genetic Algorithms, Selection, Mutation rate, Effective Fitness, Error threshold

• Type: Conference
• Year: 2006
• Where: GECCO
• Authors: Jorge Cervantes, Christopher R. Stephens
• Keywords: Error Threshold, GECCO 2006, Mutation Affects, Mutation Rate, Optimization