A Comparative Study of Large-Scale Variants of CMA-ES
The CMA-ES is one of the most powerful stochastic numerical optimizers to address difficult black-box problems. Its intrinsic time and space complexity is quadratic—limiting its applicability with increasing problem dimensionality. To circumvent this limitation, different large-scale variants of CMA...
Saved in:
Published in | Lecture notes in computer science Vol. 11101; pp. 3 - 15 |
---|---|
Main Authors | , , , , , , , |
Format | Book Chapter Conference Proceeding |
Language | English |
Published |
Cham
Springer International Publishing
Springer Science+Business Media |
Series | Lecture Notes in Computer Science |
Subjects | |
Online Access | Get more information |
Cover
Loading…
Abstract | The CMA-ES is one of the most powerful stochastic numerical optimizers to address difficult black-box problems. Its intrinsic time and space complexity is quadratic—limiting its applicability with increasing problem dimensionality. To circumvent this limitation, different large-scale variants of CMA-ES with subquadratic complexity have been proposed over the past ten years. To-date however, these variants have been tested and compared only in rather restrictive settings, due to the lack of a comprehensive large-scale testbed to assess their performance. In this context, we introduce a new large-scale testbed with dimension up to 640, implemented within the COCO benchmarking platform. We use this testbed to assess the performance of several promising variants of CMA-ES and the standard limited-memory L-BFGS. In all tested dimensions, the best CMA-ES variant solves more problems than L-BFGS for larger budgets while L-BFGS outperforms the best CMA-ES variant for smaller budgets. However, over all functions, the cumulative runtime distributions between L-BFGS and the best CMA-ES variants are close (less than a factor of 4 in high dimension).
Our results illustrate different scaling behaviors of the methods, expose a few defects of the algorithms and reveal that for dimension larger than 80, LM-CMA solves more problems than VkD-CMA while in the cumulative runtime distribution over all functions the VkD-CMA dominates LM-CMA for budgets up to 104\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$10^4$$\end{document} times dimension and for all budgets up to dimension 80. |
---|---|
AbstractList | The CMA-ES is one of the most powerful stochastic numerical optimizers to address difficult black-box problems. Its intrinsic time and space complexity is quadratic—limiting its applicability with increasing problem dimensionality. To circumvent this limitation, different large-scale variants of CMA-ES with subquadratic complexity have been proposed over the past ten years. To-date however, these variants have been tested and compared only in rather restrictive settings, due to the lack of a comprehensive large-scale testbed to assess their performance. In this context, we introduce a new large-scale testbed with dimension up to 640, implemented within the COCO benchmarking platform. We use this testbed to assess the performance of several promising variants of CMA-ES and the standard limited-memory L-BFGS. In all tested dimensions, the best CMA-ES variant solves more problems than L-BFGS for larger budgets while L-BFGS outperforms the best CMA-ES variant for smaller budgets. However, over all functions, the cumulative runtime distributions between L-BFGS and the best CMA-ES variants are close (less than a factor of 4 in high dimension).
Our results illustrate different scaling behaviors of the methods, expose a few defects of the algorithms and reveal that for dimension larger than 80, LM-CMA solves more problems than VkD-CMA while in the cumulative runtime distribution over all functions the VkD-CMA dominates LM-CMA for budgets up to 104\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$10^4$$\end{document} times dimension and for all budgets up to dimension 80. The CMA-ES is one of the most powerful stochastic numerical optimizers to address difficult black-box problems. Its intrinsic time and space complexity is quadratic-limiting its applicability with increasing problem dimensionality. To circumvent this limitation, different large-scale variants of CMA-ES with subquadratic complexity have been proposed over the past ten years. To-date however, these variants have been tested and compared only in rather restrictive settings, due to the lack of a comprehensive large-scale testbed to assess their performance. In this context, we introduce a new large-scale testbed with dimension up to 640, implemented within the COCO benchmarking platform. We use this testbed to assess the performance of several promising variants of CMA-ES and the standard limited-memory L-BFGS. In all tested dimensions, the best CMA-ES variant solves more problems than L-BFGS for larger budgets while L-BFGS outperforms the best CMA-ES variant for smaller budgets. However, over all functions, the cumulative runtime distributions between L-BFGS and the best CMA-ES variants are close (less than a factor of 4 in high dimension). Our results illustrate different scaling behaviors of the methods, expose a few defects of the algorithms and reveal that for dimension larger than 80, LM-CMA solves more problems than VkD-CMA while in the cumulative runtime distribution over all functions the VkD-CMA dominates or shows almost equal success rate with LM-CMA for budgets up to 10 4 times dimension and for all budgets up to dimension 80. |
Author | Hansen, Nikolaus Kassab, Rami Brockhoff, Dimo ElHara, Ouassim Ait Auger, Anne Semet, Yann Varelas, Konstantinos Barbaresco, Frédéric |
Author_xml | – sequence: 1 givenname: Konstantinos surname: Varelas fullname: Varelas, Konstantinos email: konstantinos.varelas@inria.fr organization: Thales LAS France SAS - Limours, Limours, France – sequence: 2 givenname: Anne surname: Auger fullname: Auger, Anne email: anne.auger@inria.fr organization: Inria, RandOpt team, CMAP, École Polytechnique, Palaiseau, France – sequence: 3 givenname: Dimo surname: Brockhoff fullname: Brockhoff, Dimo email: dimo.brockhoff@inria.fr organization: Inria, RandOpt team, CMAP, École Polytechnique, Palaiseau, France – sequence: 4 givenname: Nikolaus surname: Hansen fullname: Hansen, Nikolaus email: nikolaus.hansen@inria.fr organization: Inria, RandOpt team, CMAP, École Polytechnique, Palaiseau, France – sequence: 5 givenname: Ouassim Ait surname: ElHara fullname: ElHara, Ouassim Ait email: ouassim.elHara@inria.fr organization: Inria, RandOpt team, CMAP, École Polytechnique, Palaiseau, France – sequence: 6 givenname: Yann surname: Semet fullname: Semet, Yann email: yann.semet@thalesgroup.com organization: Thales Research Technology, Palaiseau, France – sequence: 7 givenname: Rami surname: Kassab fullname: Kassab, Rami organization: Thales LAS France SAS - Limours, Limours, France – sequence: 8 givenname: Frédéric surname: Barbaresco fullname: Barbaresco, Frédéric organization: Thales LAS France SAS - Limours, Limours, France |
BackLink | https://inria.hal.science/hal-01881454$$DView record in HAL |
BookMark | eNqFkMtOwzAQRQ0UiVL6BWyyZWEYP_LwMoqAIgWxKLC1polTAmlcOWlR_x63QUJsYGXp-B6P556TUWtbQ8glg2sGEN-oOKGCCqaoUjwUlGt2RKaeCs8OiB-TMYsYo0JIdfLrjocjMgYBnKpYijMy7bp3AOCQqATUmIRpkNnVGh329dYE835T7gJbBTm6paHzAhsTvKKrse27Pc8eU3o7vyCnFTadmX6fE_Jyd_uczWj-dP-QpTktBI96WhouSgZhHCHjpoqAS1bFURwuRBRKI8pFEiWSV1xhWUihoOTg0yVCZbDy35wQOby7ade4-8Sm0WtXr9DtNAO9b0f7XbXQflt9qEL7drx2NWhv-CNYrPUszfWeAUsSJkO55T6rhmzhbNc5U-mi7n0Ztu0d1s0_c9jgdn5IuzROL6z96P50vgB6cIDU |
CitedBy_id | crossref_primary_10_1002_tee_23201 crossref_primary_10_1109_TBIOM_2022_3223738 crossref_primary_10_3390_pr11113188 crossref_primary_10_1109_ACCESS_2023_3254896 crossref_primary_10_1162_evco_a_00275 crossref_primary_10_1109_MAES_2021_3052312 crossref_primary_10_1080_10556788_2020_1808977 crossref_primary_10_1162_evco_a_00325 crossref_primary_10_1109_TEMC_2020_2964059 crossref_primary_10_1007_s10489_022_04265_x crossref_primary_10_1016_j_tafmec_2023_104077 crossref_primary_10_1109_LRA_2023_3313012 crossref_primary_10_1109_TEVC_2023_3266955 crossref_primary_10_2139_ssrn_3307212 crossref_primary_10_2139_ssrn_3292762 crossref_primary_10_2139_ssrn_3365453 crossref_primary_10_1007_s11390_021_1213_3 |
ContentType | Book Chapter Conference Proceeding |
Copyright | Springer Nature Switzerland AG 2018 Distributed under a Creative Commons Attribution 4.0 International License |
Copyright_xml | – notice: Springer Nature Switzerland AG 2018 – notice: Distributed under a Creative Commons Attribution 4.0 International License |
DBID | 1XC VOOES UNPAY |
DOI | 10.1007/978-3-319-99253-2_1 |
DatabaseName | Hyper Article en Ligne (HAL) Hyper Article en Ligne (HAL) (Open Access) Unpaywall |
DatabaseTitleList | |
DeliveryMethod | no_fulltext_linktorsrc |
Discipline | Engineering Computer Science Mathematics |
EISBN | 9783319992532 3319992538 |
EISSN | 1611-3349 |
Editor | Whitley, Darrell Fonseca, Carlos M. Machado, Penousal Paquete, Luís Lourenço, Nuno Auger, Anne |
Editor_xml | – sequence: 1 givenname: Anne surname: Auger fullname: Auger, Anne email: anne.auger@inria.fr – sequence: 2 givenname: Carlos M. surname: Fonseca fullname: Fonseca, Carlos M. email: cmfonsec@dei.uc.pt – sequence: 3 givenname: Nuno surname: Lourenço fullname: Lourenço, Nuno email: naml@dei.uc.pt – sequence: 4 givenname: Penousal surname: Machado fullname: Machado, Penousal email: machado@dei.uc.pt – sequence: 5 givenname: Luís surname: Paquete fullname: Paquete, Luís email: paquete@dei.uc.pt – sequence: 6 givenname: Darrell surname: Whitley fullname: Whitley, Darrell email: whitley@cs.colostate.edu |
EndPage | 15 |
ExternalDocumentID | 10_1007_978_3_319_99253_2_1 oai_HAL_hal_01881454v2 n/a |
GroupedDBID | -DT -GH -~X 1SB 29L 2HA 2HV 5QI 875 AASHB ABMNI ACGFS ADCXD AEFIE ALMA_UNASSIGNED_HOLDINGS EJD F5P FEDTE HVGLF LAS LDH P2P RIG RNI RSU SVGTG VI1 ~02 1XC VOOES UNPAY |
ID | FETCH-LOGICAL-c326t-de23d10576a12ef60241f7675b3654e3db86842f29adc4390d20057da0feaf000 |
ISBN | 9783319992525 331999252X |
ISSN | 0302-9743 1611-3349 |
IngestDate | Wed Aug 28 01:18:23 EDT 2024 Tue Oct 15 15:49:25 EDT 2024 Tue Jul 23 14:21:24 EDT 2024 Tue Oct 01 19:09:30 EDT 2024 |
IsDoiOpenAccess | true |
IsOpenAccess | true |
IsPeerReviewed | true |
IsScholarly | true |
Language | English |
License | Distributed under a Creative Commons Attribution 4.0 International License: http://creativecommons.org/licenses/by/4.0 |
LinkModel | DirectLink |
MeetingName | PPSN XV 2018 - 15th International Conference on Parallel Problem Solving from Nature |
MergedId | FETCHMERGED-LOGICAL-c326t-de23d10576a12ef60241f7675b3654e3db86842f29adc4390d20057da0feaf000 |
ORCID | 0000-0001-7788-4906 0000-0003-3664-3609 |
OpenAccessLink | https://inria.hal.science/hal-01881454 |
PageCount | 13 |
ParticipantIDs | unpaywall_primary_10_1007_978_3_319_99253_2_1 hal_primary_oai_HAL_hal_01881454v2 crossref_citationtrail_10_1007_978_3_319_99253_2_1 springer_books_10_1007_978_3_319_99253_2_1 |
PublicationPlace | Cham |
PublicationPlace_xml | – name: Cham |
PublicationSeriesSubtitle | Theoretical Computer Science and General Issues |
PublicationSeriesTitle | Lecture Notes in Computer Science |
PublicationSeriesTitleAlternate | Lect.Notes Computer |
PublicationSubtitle | 15th International Conference, Coimbra, Portugal, September 8–12, 2018, Proceedings, Part I |
PublicationTitle | Lecture notes in computer science |
Publisher | Springer International Publishing Springer Science+Business Media |
Publisher_xml | – name: Springer International Publishing – name: Springer Science+Business Media |
RelatedPersons | Kleinberg, Jon M. Mattern, Friedemann Naor, Moni Mitchell, John C. Terzopoulos, Demetri Steffen, Bernhard Pandu Rangan, C. Kanade, Takeo Kittler, Josef Weikum, Gerhard Hutchison, David Tygar, Doug |
RelatedPersons_xml | – sequence: 1 givenname: David surname: Hutchison fullname: Hutchison, David organization: Lancaster University, Lancaster, United Kingdom – sequence: 2 givenname: Takeo surname: Kanade fullname: Kanade, Takeo organization: Carnegie Mellon University, Pittsburgh, USA – sequence: 3 givenname: Josef surname: Kittler fullname: Kittler, Josef organization: University of Surrey, Guildford, United Kingdom – sequence: 4 givenname: Jon M. surname: Kleinberg fullname: Kleinberg, Jon M. organization: Cornell University, Ithaca, USA – sequence: 5 givenname: Friedemann surname: Mattern fullname: Mattern, Friedemann organization: ETH Zurich, Zurich, Switzerland – sequence: 6 givenname: John C. surname: Mitchell fullname: Mitchell, John C. organization: Stanford University, Stanford, USA – sequence: 7 givenname: Moni surname: Naor fullname: Naor, Moni organization: Dept Applied Math & Computer Science, Weizmann Institute of Science, Rehovot, Israel – sequence: 8 givenname: C. surname: Pandu Rangan fullname: Pandu Rangan, C. organization: Indian Institute of Technology Madras, Chennai, India – sequence: 9 givenname: Bernhard surname: Steffen fullname: Steffen, Bernhard organization: TU Dortmund University, Dortmund, Germany – sequence: 10 givenname: Demetri surname: Terzopoulos fullname: Terzopoulos, Demetri organization: University of California, Los Angeles, USA – sequence: 11 givenname: Doug surname: Tygar fullname: Tygar, Doug organization: University of California, Berkeley, USA – sequence: 12 givenname: Gerhard surname: Weikum fullname: Weikum, Gerhard organization: Max Planck Institute for Informatics, Saarbrücken, Germany |
SSID | ssj0002089809 ssj0002792 |
Score | 2.371786 |
Snippet | The CMA-ES is one of the most powerful stochastic numerical optimizers to address difficult black-box problems. Its intrinsic time and space complexity is... |
SourceID | unpaywall hal crossref springer |
SourceType | Open Access Repository Index Database Publisher |
StartPage | 3 |
SubjectTerms | Computer Science Mathematics Neural and Evolutionary Computing Optimization and Control |
Title | A Comparative Study of Large-Scale Variants of CMA-ES |
URI | http://link.springer.com/10.1007/978-3-319-99253-2_1 https://inria.hal.science/hal-01881454 https://inria.hal.science/hal-01881454/document |
Volume | 11101 |
hasFullText | |
inHoldings | |
isFullTextHit | |
isPrint | |
link | http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwjV1NSwMxEA1WD9KLWitWFIJ4EqK72ezXsRRLkVYErfQW8kkPZVtoVfrvnWl3t3op3pawTJYXMm9mM3lDyF1uvA8TnjKuM8OQchn4vJQFgQO68kKLbbXFSzIYi-dJPCkLZDd3YQrA_WEKcWfJAI_wDPluloUiFg3SgJC5Fg_a6sPi_bAcZuBxxLgM_xBMY4rljfUZZ5McfxYLtf5Ws9kvGumfkvbugh19ranjjBy4okVOqg4LtNxwLdIc1aqqy3MSd2lvJ9ZNsQRwTeeeDrGamy0BbUc_IPfF0hYc74267OmtTcb9p_fegJVtD5iBWGrFrOORxfa7iQq58wmwaOhRc0VHSSxcZHWGh2ee58oaiCcCi3-GUqsC75QHF3dBDot54S4JVcJm1mpwaqkVxoYq1zZIPcRwOuUm5B3CK6ykKTXBsTXFTFZqxgCwjCQALDcASwC4Q24BVbnYqmFI1KcedIcSx6p1-gLL9xXoEtOF5X6LrF6W2u6e96_-8wXX5MjDDnY3EDSs9A_l9rwA |
link.rule.ids | 230,310,311,785,786,891,25170 |
openUrl | ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=bookitem&rft.title=Parallel+Problem+Solving+from+Nature+%E2%80%93+PPSN+XV&rft.au=Varelas%2C+Konstantinos&rft.au=Auger%2C+Anne&rft.au=Brockhoff%2C+Dimo&rft.au=Hansen%2C+Nikolaus&rft.atitle=A+Comparative+Study+of+Large-Scale+Variants+of+CMA-ES&rft.series=Lecture+Notes+in+Computer+Science&rft.pub=Springer+International+Publishing&rft.isbn=9783319992525&rft.issn=0302-9743&rft.eissn=1611-3349&rft.spage=3&rft.epage=15&rft_id=info:doi/10.1007%2F978-3-319-99253-2_1 |
thumbnail_l | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=0302-9743&client=summon |
thumbnail_m | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=0302-9743&client=summon |
thumbnail_s | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=0302-9743&client=summon |