Performance analysis of niching algorithms based on derandomized-ES variants

A survey of niching algorithms, based on 5 variants of derandomized Evolution Strategies (ES), is introduced. This set of niching algorithms, ranging from the very first derandomized approach to self-adaptation of ES to the sophisticated (1 +, λ) Covariance Matrix Adaptation (CMA), is applied to multimodal continuous theoretical test functions, of different levels of difficulty and various dimensions, and compared with the MPR performance analysis tool. While characterizing the performance of the different derandomized variants in the context of niching, some conclusions concerning the niching formation process of the different mechanisms are drawn, and the hypothesis of a tradeoff between learning time and niching acceleration is numerically confirmed. Niching with (1 + λ)-CMA core mechanism is shown to experimentally outperform all the other variants. Some theoretical arguments supporting the advantage of a plus-strategy for niching are discussed. Categories and Subject Desc...

• Type: Conference
• Year: 2007
• Where: GECCO
• Authors: Ofer M. Shir, Thomas Bäck
• Keywords: Different Derandomized Variants, Evolution Strategies, GECCO 2007, first Derandomized Approach