Learning Step-Size Adaptation in CMA-ES

Publikation: Beitrag in Buch/Bericht/Sammelwerk/KonferenzbandAufsatz in KonferenzbandForschungPeer-Review

Autoren

  • Gresa Shala
  • André Biedenkapp
  • Noor Awad
  • Steven Adriaensen
  • Marius Lindauer
  • Frank Hutter

Externe Organisationen

  • Albert-Ludwigs-Universität Freiburg
  • Bosch Center for Artificial Intelligence (BCAI)
Forschungs-netzwerk anzeigen

Details

OriginalspracheEnglisch
Titel des SammelwerksParallel Problem Solving from Nature – PPSN XVI
Untertitel16th International Conference, PPSN 2020, Leiden, The Netherlands, September 5-9, 2020, Proceedings, Part I
Herausgeber/-innenThomas Bäck, Mike Preuss, André Deutz, Michael Emmerich, Hao Wang, Carola Doerr, Heike Trautmann
ErscheinungsortCham
Seiten691-706
Seitenumfang16
ISBN (elektronisch)978-3-030-58112-1
PublikationsstatusVeröffentlicht - 2020
Veranstaltung16th International Conference on Parallel Problem Solving from Nature, PPSN 2020 - Leiden, Niederlande
Dauer: 5 Sept. 20209 Sept. 2020

Publikationsreihe

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Band12269
ISSN (Print)0302-9743
ISSN (elektronisch)1611-3349

Abstract

An algorithm’s parameter setting often affects its ability to solve a given problem, e.g., population-size, mutation-rate or crossover-rate of an evolutionary algorithm. Furthermore, some parameters have to be adjusted dynamically, such as lowering the mutation-strength over time. While hand-crafted heuristics offer a way to fine-tune and dynamically configure these parameters, their design is tedious, time-consuming and typically involves analyzing the algorithm’s behavior on simple problems that may not be representative for those that arise in practice. In this paper, we show that formulating dynamic algorithm configuration as a reinforcement learning problem allows us to automatically learn policies that can dynamically configure the mutation step-size parameter of Covariance Matrix Adaptation Evolution Strategy (CMA-ES). We evaluate our approach on a wide range of black-box optimization problems, and show that (i) learning step-size policies has the potential to improve the performance of CMA-ES; (ii) learned step-size policies can outperform the default Cumulative Step-Size Adaptation of CMA-ES; and transferring the policies to (iii) different function classes and to (iv) higher dimensions is also possible.

ASJC Scopus Sachgebiete

Zitieren

Learning Step-Size Adaptation in CMA-ES. / Shala, Gresa; Biedenkapp, André; Awad, Noor et al.
Parallel Problem Solving from Nature – PPSN XVI: 16th International Conference, PPSN 2020, Leiden, The Netherlands, September 5-9, 2020, Proceedings, Part I. Hrsg. / Thomas Bäck; Mike Preuss; André Deutz; Michael Emmerich; Hao Wang; Carola Doerr; Heike Trautmann. Cham, 2020. S. 691-706 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Band 12269).

Publikation: Beitrag in Buch/Bericht/Sammelwerk/KonferenzbandAufsatz in KonferenzbandForschungPeer-Review

Shala, G, Biedenkapp, A, Awad, N, Adriaensen, S, Lindauer, M & Hutter, F 2020, Learning Step-Size Adaptation in CMA-ES. in T Bäck, M Preuss, A Deutz, M Emmerich, H Wang, C Doerr & H Trautmann (Hrsg.), Parallel Problem Solving from Nature – PPSN XVI: 16th International Conference, PPSN 2020, Leiden, The Netherlands, September 5-9, 2020, Proceedings, Part I. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Bd. 12269, Cham, S. 691-706, 16th International Conference on Parallel Problem Solving from Nature, PPSN 2020, Leiden, Niederlande, 5 Sept. 2020. https://doi.org/10.1007/978-3-030-58112-1_48
Shala, G., Biedenkapp, A., Awad, N., Adriaensen, S., Lindauer, M., & Hutter, F. (2020). Learning Step-Size Adaptation in CMA-ES. In T. Bäck, M. Preuss, A. Deutz, M. Emmerich, H. Wang, C. Doerr, & H. Trautmann (Hrsg.), Parallel Problem Solving from Nature – PPSN XVI: 16th International Conference, PPSN 2020, Leiden, The Netherlands, September 5-9, 2020, Proceedings, Part I (S. 691-706). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Band 12269).. https://doi.org/10.1007/978-3-030-58112-1_48
Shala G, Biedenkapp A, Awad N, Adriaensen S, Lindauer M, Hutter F. Learning Step-Size Adaptation in CMA-ES. in Bäck T, Preuss M, Deutz A, Emmerich M, Wang H, Doerr C, Trautmann H, Hrsg., Parallel Problem Solving from Nature – PPSN XVI: 16th International Conference, PPSN 2020, Leiden, The Netherlands, September 5-9, 2020, Proceedings, Part I. Cham. 2020. S. 691-706. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). Epub 2020 Aug 31. doi: 10.1007/978-3-030-58112-1_48
Shala, Gresa ; Biedenkapp, André ; Awad, Noor et al. / Learning Step-Size Adaptation in CMA-ES. Parallel Problem Solving from Nature – PPSN XVI: 16th International Conference, PPSN 2020, Leiden, The Netherlands, September 5-9, 2020, Proceedings, Part I. Hrsg. / Thomas Bäck ; Mike Preuss ; André Deutz ; Michael Emmerich ; Hao Wang ; Carola Doerr ; Heike Trautmann. Cham, 2020. S. 691-706 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
Download
@inproceedings{c2ab955cefa549a5a965fe626d5fe316,
title = "Learning Step-Size Adaptation in CMA-ES",
abstract = "An algorithm{\textquoteright}s parameter setting often affects its ability to solve a given problem, e.g., population-size, mutation-rate or crossover-rate of an evolutionary algorithm. Furthermore, some parameters have to be adjusted dynamically, such as lowering the mutation-strength over time. While hand-crafted heuristics offer a way to fine-tune and dynamically configure these parameters, their design is tedious, time-consuming and typically involves analyzing the algorithm{\textquoteright}s behavior on simple problems that may not be representative for those that arise in practice. In this paper, we show that formulating dynamic algorithm configuration as a reinforcement learning problem allows us to automatically learn policies that can dynamically configure the mutation step-size parameter of Covariance Matrix Adaptation Evolution Strategy (CMA-ES). We evaluate our approach on a wide range of black-box optimization problems, and show that (i) learning step-size policies has the potential to improve the performance of CMA-ES; (ii) learned step-size policies can outperform the default Cumulative Step-Size Adaptation of CMA-ES; and transferring the policies to (iii) different function classes and to (iv) higher dimensions is also possible.",
keywords = "Algorithm configuration, Evolutionary algorithms, Reinforcement learning",
author = "Gresa Shala and Andr{\'e} Biedenkapp and Noor Awad and Steven Adriaensen and Marius Lindauer and Frank Hutter",
note = "Funding Information: The authors acknowledge funding by the Robert Bosch GmbH.; 16th International Conference on Parallel Problem Solving from Nature, PPSN 2020 ; Conference date: 05-09-2020 Through 09-09-2020",
year = "2020",
doi = "10.1007/978-3-030-58112-1_48",
language = "English",
isbn = "9783030581114",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
pages = "691--706",
editor = "Thomas B{\"a}ck and Mike Preuss and Andr{\'e} Deutz and Michael Emmerich and Hao Wang and Carola Doerr and Heike Trautmann",
booktitle = "Parallel Problem Solving from Nature – PPSN XVI",

}

Download

TY - GEN

T1 - Learning Step-Size Adaptation in CMA-ES

AU - Shala, Gresa

AU - Biedenkapp, André

AU - Awad, Noor

AU - Adriaensen, Steven

AU - Lindauer, Marius

AU - Hutter, Frank

N1 - Funding Information: The authors acknowledge funding by the Robert Bosch GmbH.

PY - 2020

Y1 - 2020

N2 - An algorithm’s parameter setting often affects its ability to solve a given problem, e.g., population-size, mutation-rate or crossover-rate of an evolutionary algorithm. Furthermore, some parameters have to be adjusted dynamically, such as lowering the mutation-strength over time. While hand-crafted heuristics offer a way to fine-tune and dynamically configure these parameters, their design is tedious, time-consuming and typically involves analyzing the algorithm’s behavior on simple problems that may not be representative for those that arise in practice. In this paper, we show that formulating dynamic algorithm configuration as a reinforcement learning problem allows us to automatically learn policies that can dynamically configure the mutation step-size parameter of Covariance Matrix Adaptation Evolution Strategy (CMA-ES). We evaluate our approach on a wide range of black-box optimization problems, and show that (i) learning step-size policies has the potential to improve the performance of CMA-ES; (ii) learned step-size policies can outperform the default Cumulative Step-Size Adaptation of CMA-ES; and transferring the policies to (iii) different function classes and to (iv) higher dimensions is also possible.

AB - An algorithm’s parameter setting often affects its ability to solve a given problem, e.g., population-size, mutation-rate or crossover-rate of an evolutionary algorithm. Furthermore, some parameters have to be adjusted dynamically, such as lowering the mutation-strength over time. While hand-crafted heuristics offer a way to fine-tune and dynamically configure these parameters, their design is tedious, time-consuming and typically involves analyzing the algorithm’s behavior on simple problems that may not be representative for those that arise in practice. In this paper, we show that formulating dynamic algorithm configuration as a reinforcement learning problem allows us to automatically learn policies that can dynamically configure the mutation step-size parameter of Covariance Matrix Adaptation Evolution Strategy (CMA-ES). We evaluate our approach on a wide range of black-box optimization problems, and show that (i) learning step-size policies has the potential to improve the performance of CMA-ES; (ii) learned step-size policies can outperform the default Cumulative Step-Size Adaptation of CMA-ES; and transferring the policies to (iii) different function classes and to (iv) higher dimensions is also possible.

KW - Algorithm configuration

KW - Evolutionary algorithms

KW - Reinforcement learning

UR - http://www.scopus.com/inward/record.url?scp=85091295368&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-58112-1_48

DO - 10.1007/978-3-030-58112-1_48

M3 - Conference contribution

AN - SCOPUS:85091295368

SN - 9783030581114

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 691

EP - 706

BT - Parallel Problem Solving from Nature – PPSN XVI

A2 - Bäck, Thomas

A2 - Preuss, Mike

A2 - Deutz, André

A2 - Emmerich, Michael

A2 - Wang, Hao

A2 - Doerr, Carola

A2 - Trautmann, Heike

CY - Cham

T2 - 16th International Conference on Parallel Problem Solving from Nature, PPSN 2020

Y2 - 5 September 2020 through 9 September 2020

ER -

Von denselben Autoren