Large-Scale Evolution Strategy Based on Search Direction Adaptation

被引:21
作者
He, Xiaoyu [1 ,2 ]
Zhou, Yuren [1 ,2 ]
Chen, Zefeng [1 ,2 ]
Zhang, Jun [3 ]
Chen, Wei-Neng [4 ]
机构
[1] Sun Yat Sen Univ, Sch Data & Comp Sci, Guangzhou 510006, Peoples R China
[2] Sun Yat Sen Univ, Key Lab Machine Intelligence & Adv Comp, Minist Educ, Guangzhou 510006, Peoples R China
[3] Univ Victoria, Melbourne, Vic 8001, Australia
[4] South China Univ Technol, Sch Comp Sci & Engn, Guangzhou 510006, Peoples R China
基金
中国国家自然科学基金;
关键词
Covariance matrices; Principal component analysis; Complexity theory; Optimization; Linear programming; Correlation; Adaptation models; Evolution strategy; large-scale optimization; search direction adaptation;
D O I
10.1109/TCYB.2019.2928563
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
The covariance matrix adaptation evolution strategy (CMA-ES) is a powerful evolutionary algorithm for single-objective real-valued optimization. However, the time and space complexity may preclude its use in high-dimensional decision space. Recent studies suggest that putting sparse or low-rank constraints on the structure of the covariance matrix can improve the efficiency of CMA-ES in handling large-scale problems. Following this idea, this paper proposes a search direction adaptation evolution strategy (SDA-ES) which achieves linear time and space complexity. SDA-ES models the covariance matrix with an identity matrix and multiple search directions, and uses a heuristic to update the search directions in a way similar to the principal component analysis. We also generalize the traditional 1/5th success rule to adapt the mutation strength which exhibits the derandomization property. Numerical comparisons with nine state-of-the-art algorithms are carried out on 31 test problems. The experimental results have shown that SDA-ES is invariant under search-space rotational transformations, and is scalable with respect to the number of variables. It also achieves competitive performance on generic black-box problems, demonstrating its effectiveness in keeping a good tradeoff between solution quality and computational efficiency.
引用
收藏
页码:1651 / 1665
页数:15
相关论文
共 52 条
[1]   Projection-Based Restricted Covariance Matrix Adaptation for High Dimension [J].
Akimoto, Youhei ;
Hansen, Nikolaus .
GECCO'16: PROCEEDINGS OF THE 2016 GENETIC AND EVOLUTIONARY COMPUTATION CONFERENCE, 2016, :197-204
[2]   Comparison-Based Natural Gradient Optimization in High Dimension [J].
Akimoto, Youhei ;
Auger, Anne ;
Hansen, Nikolaus .
GECCO'14: PROCEEDINGS OF THE 2014 GENETIC AND EVOLUTIONARY COMPUTATION CONFERENCE, 2014, :373-380
[3]   Natural gradient works efficiently in learning [J].
Amari, S .
NEURAL COMPUTATION, 1998, 10 (02) :251-276
[4]   Weighted multirecombination evolution strategies [J].
Arnold, Dirk V. .
THEORETICAL COMPUTER SCIENCE, 2006, 361 (01) :18-37
[5]   Performance analysis of evolutionary optimization with cumulative step length adaptation [J].
Arnold, DV ;
Beyer, HG .
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 2004, 49 (04) :617-622
[6]  
Auger A, 2005, IEEE C EVOL COMPUTAT, P1769
[7]  
Beyer H.-G., 2001, The Theory of Evolution Strategies
[8]   Simplify Your Covariance Matrix Adaptation Evolution Strategy [J].
Beyer, Hans-Georg ;
Sendhoff, Bernhard .
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION, 2017, 21 (05) :746-759
[9]   The Dynamics of Cumulative Step Size Adaptation on the Ellipsoid Model [J].
Beyer, Hans-Georg ;
Hellwig, Michael .
EVOLUTIONARY COMPUTATION, 2016, 24 (01) :25-57
[10]   Convergence Analysis of Evolutionary Algorithms That Are Based on the Paradigm of Information Geometry [J].
Beyer, Hans-Georg .
EVOLUTIONARY COMPUTATION, 2014, 22 (04) :679-709