Associate Professor Tsung-Che Chiang

Department of Computer Science and Information Engineering
National Taiwan Normal University

Tel: +886-2-77346692    Fax: +886-2-29322378    Email

Jia-Fong Yeh, Ting-Yu Chen, and Tsung-Che Chiang, Modified L-SHADE for single objective real-parameter optimization, Proc. of IEEE Congress on Evolutionary Computation (CEC), pp. 373-378, Willington, New Zealand, Jun. 11-13, 2019.

• Full text: Official version; Unedited version / Presentation Slides
• Source code: GitHub

Abstract

In this paper we address single objective real parameter optimization by using differential evolution (DE). L-SHADE is a well-known DE with success history-based adaptation and linear population size reduction. We propose a modified L-SHADE (mL-SHADE), in which three modifications are made: (1) removal of the terminal value, (2) addition of polynomial mutation, and (3) proposal of a memory perturbation mechanism. Performance of the proposed mL-SHADE is verified by using ten benchmark functions in the CEC2019 100-Digit Challenge. The results show that mL-SHADE achieves a higher score than seven state-of-the-art adaptive evolutionary algorithms.

CEC2019 Competition Results

The following two tables are taken from the final report of the competition, which should be available on the official site of the CEC2019 Competition on 100-Digit Challenge and Numerical Optimization (github). Our mL-SHADE algorithm is ranked the 11th place among 18 entries in terms of the competiton score. However, no algorithm can dominate our algorithm when solution quality (score) and computational effort (number of function evaluations, nFE) are considered simultaneously. The maximum allowed function evaluations of HyDE-DF and ESHADE-USM in the final report seem not to be consistent with the data in the original papers. We drew the score-versus-nFE figure based on our understanding of their nFE.