MATE: A Model-Based Algorithm Tuning Engine: A Proof of Concept Towards Transparent Feature-Dependent Parameter Tuning Using Symbolic Regression

Mohamed El Yafrani; Marcella Scoczynski; Inkyung Sung; Markus Wagner; Carola Doerr; Peter Nielsen

doi:10.1007/978-3-030-72904-2_4

MATE: A Model-Based Algorithm Tuning Engine: A Proof of Concept Towards Transparent Feature-Dependent Parameter Tuning Using Symbolic Regression

Mohamed El Yafrani^*, Marcella Scoczynski, Inkyung Sung, Markus Wagner, Carola Doerr, Peter Nielsen

^*Corresponding author for this work

Research output: Contribution to book/anthology/report/conference proceeding › Article in proceeding › Research › peer-review

2 Citations (Scopus)

Abstract

In this paper, we introduce a Model-based Algorithm Tuning Engine, namely MATE, where the parameters of an algorithm are represented as expressions of the features of a target optimisation problem. In contrast to most static (feature-independent) algorithm tuning engines such as irace and SPOT, our approach aims to derive the best parameter configuration of a given algorithm for a specific problem, exploiting the relationships between the algorithm parameters and the features of the problem. We formulate the problem of finding the relationships between the parameters and the problem features as a symbolic regression problem and we use genetic programming to extract these expressions in a human-readable form. For the evaluation, we apply our approach to the configuration of the (1 + 1) EA and RLS algorithms for the OneMax, LeadingOnes, BinValue and Jump optimisation problems, where the theoretically optimal algorithm parameters to the problems are available as functions of the features of the problems. Our study shows that the found relationships typically comply with known theoretical results – this demonstrates (1) the potential of model-based parameter tuning as an alternative to existing static algorithm tuning engines, and (2) its potential to discover relationships between algorithm performance and instance features in human-readable form.

Original language	English
Title of host publication	Evolutionary Computation in Combinatorial Optimization - 21st European Conference, EvoCOP 2021, Held as Part of EvoStar 2021, Proceedings
Editors	Christine Zarges, Sébastien Verel
Number of pages	17
Publisher	Springer Science+Business Media
Publication date	2021
Pages	51-67
ISBN (Print)	9783030729035
DOIs	https://doi.org/10.1007/978-3-030-72904-2_4
Publication status	Published - 2021
Event	21st European Conference on Evolutionary Computation in Combinatorial Optimization, EvoCOP 2021 Held as Part of EvoStar 2021 - Virtual, Online Duration: 7 Apr 2021 → 9 Apr 2021

Conference

Conference	21st European Conference on Evolutionary Computation in Combinatorial Optimization, EvoCOP 2021 Held as Part of EvoStar 2021
City	Virtual, Online
Period	07/04/2021 → 09/04/2021

Series	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12692 LNCS
ISSN	0302-9743

Bibliographical note

Funding Information:
M. Martins acknowledges CNPq (Brazil Government). M. Wagner acknowledges the ARC Discovery Early Career Researcher Award DE160100850. C. Doerr acknowledges support from the Paris Ile-de-France Region. Experiments were performed on the AAU?s CLAUDIA compute cloud platform.

Publisher Copyright:
© 2021, Springer Nature Switzerland AG.

Keywords

Genetic programming
Model-based tuning
Parameter tuning

Access to Document

10.1007/978-3-030-72904-2_4

https://arxiv.org/pdf/2004.12750.pdf

AUB Link

Search for the material in Aalborg University Library's search engine

Cite this

El Yafrani, M., Scoczynski, M., Sung, I., Wagner, M., Doerr, C., & Nielsen, P. (2021). MATE: A Model-Based Algorithm Tuning Engine: A Proof of Concept Towards Transparent Feature-Dependent Parameter Tuning Using Symbolic Regression. In C. Zarges, & S. Verel (Eds.), Evolutionary Computation in Combinatorial Optimization - 21st European Conference, EvoCOP 2021, Held as Part of EvoStar 2021, Proceedings (pp. 51-67). Springer Science+Business Media. https://doi.org/10.1007/978-3-030-72904-2_4

El Yafrani, Mohamed ; Scoczynski, Marcella ; Sung, Inkyung et al. / MATE: A Model-Based Algorithm Tuning Engine : A Proof of Concept Towards Transparent Feature-Dependent Parameter Tuning Using Symbolic Regression. Evolutionary Computation in Combinatorial Optimization - 21st European Conference, EvoCOP 2021, Held as Part of EvoStar 2021, Proceedings. editor / Christine Zarges ; Sébastien Verel. Springer Science+Business Media, 2021. pp. 51-67 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 12692 LNCS).

@inproceedings{828b56d46417436b8f1170bcfde82e5e,

title = "MATE: A Model-Based Algorithm Tuning Engine: A Proof of Concept Towards Transparent Feature-Dependent Parameter Tuning Using Symbolic Regression",

abstract = "In this paper, we introduce a Model-based Algorithm Tuning Engine, namely MATE, where the parameters of an algorithm are represented as expressions of the features of a target optimisation problem. In contrast to most static (feature-independent) algorithm tuning engines such as irace and SPOT, our approach aims to derive the best parameter configuration of a given algorithm for a specific problem, exploiting the relationships between the algorithm parameters and the features of the problem. We formulate the problem of finding the relationships between the parameters and the problem features as a symbolic regression problem and we use genetic programming to extract these expressions in a human-readable form. For the evaluation, we apply our approach to the configuration of the (1 + 1) EA and RLS algorithms for the OneMax, LeadingOnes, BinValue and Jump optimisation problems, where the theoretically optimal algorithm parameters to the problems are available as functions of the features of the problems. Our study shows that the found relationships typically comply with known theoretical results – this demonstrates (1) the potential of model-based parameter tuning as an alternative to existing static algorithm tuning engines, and (2) its potential to discover relationships between algorithm performance and instance features in human-readable form.",

keywords = "Genetic programming, Model-based tuning, Parameter tuning",

author = "{El Yafrani}, Mohamed and Marcella Scoczynski and Inkyung Sung and Markus Wagner and Carola Doerr and Peter Nielsen",

note = "Funding Information: M. Martins acknowledges CNPq (Brazil Government). M. Wagner acknowledges the ARC Discovery Early Career Researcher Award DE160100850. C. Doerr acknowledges support from the Paris Ile-de-France Region. Experiments were performed on the AAU?s CLAUDIA compute cloud platform. Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; 21st European Conference on Evolutionary Computation in Combinatorial Optimization, EvoCOP 2021 Held as Part of EvoStar 2021 ; Conference date: 07-04-2021 Through 09-04-2021",

year = "2021",

doi = "10.1007/978-3-030-72904-2_4",

language = "English",

isbn = "9783030729035",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science+Business Media",

pages = "51--67",

editor = "Christine Zarges and S{\'e}bastien Verel",

booktitle = "Evolutionary Computation in Combinatorial Optimization - 21st European Conference, EvoCOP 2021, Held as Part of EvoStar 2021, Proceedings",

address = "United States",

}

El Yafrani, M, Scoczynski, M, Sung, I, Wagner, M, Doerr, C & Nielsen, P 2021, MATE: A Model-Based Algorithm Tuning Engine: A Proof of Concept Towards Transparent Feature-Dependent Parameter Tuning Using Symbolic Regression. in C Zarges & S Verel (eds), Evolutionary Computation in Combinatorial Optimization - 21st European Conference, EvoCOP 2021, Held as Part of EvoStar 2021, Proceedings. Springer Science+Business Media, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12692 LNCS, pp. 51-67, 21st European Conference on Evolutionary Computation in Combinatorial Optimization, EvoCOP 2021 Held as Part of EvoStar 2021, Virtual, Online, 07/04/2021. https://doi.org/10.1007/978-3-030-72904-2_4

MATE: A Model-Based Algorithm Tuning Engine: A Proof of Concept Towards Transparent Feature-Dependent Parameter Tuning Using Symbolic Regression. / El Yafrani, Mohamed; Scoczynski, Marcella; Sung, Inkyung et al.
Evolutionary Computation in Combinatorial Optimization - 21st European Conference, EvoCOP 2021, Held as Part of EvoStar 2021, Proceedings. ed. / Christine Zarges; Sébastien Verel. Springer Science+Business Media, 2021. p. 51-67 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 12692 LNCS).

Research output: Contribution to book/anthology/report/conference proceeding › Article in proceeding › Research › peer-review

TY - GEN

T1 - MATE: A Model-Based Algorithm Tuning Engine

T2 - 21st European Conference on Evolutionary Computation in Combinatorial Optimization, EvoCOP 2021 Held as Part of EvoStar 2021

AU - El Yafrani, Mohamed

AU - Scoczynski, Marcella

AU - Sung, Inkyung

AU - Wagner, Markus

AU - Doerr, Carola

AU - Nielsen, Peter

N1 - Funding Information: M. Martins acknowledges CNPq (Brazil Government). M. Wagner acknowledges the ARC Discovery Early Career Researcher Award DE160100850. C. Doerr acknowledges support from the Paris Ile-de-France Region. Experiments were performed on the AAU?s CLAUDIA compute cloud platform. Publisher Copyright: © 2021, Springer Nature Switzerland AG.

PY - 2021

Y1 - 2021

N2 - In this paper, we introduce a Model-based Algorithm Tuning Engine, namely MATE, where the parameters of an algorithm are represented as expressions of the features of a target optimisation problem. In contrast to most static (feature-independent) algorithm tuning engines such as irace and SPOT, our approach aims to derive the best parameter configuration of a given algorithm for a specific problem, exploiting the relationships between the algorithm parameters and the features of the problem. We formulate the problem of finding the relationships between the parameters and the problem features as a symbolic regression problem and we use genetic programming to extract these expressions in a human-readable form. For the evaluation, we apply our approach to the configuration of the (1 + 1) EA and RLS algorithms for the OneMax, LeadingOnes, BinValue and Jump optimisation problems, where the theoretically optimal algorithm parameters to the problems are available as functions of the features of the problems. Our study shows that the found relationships typically comply with known theoretical results – this demonstrates (1) the potential of model-based parameter tuning as an alternative to existing static algorithm tuning engines, and (2) its potential to discover relationships between algorithm performance and instance features in human-readable form.

AB - In this paper, we introduce a Model-based Algorithm Tuning Engine, namely MATE, where the parameters of an algorithm are represented as expressions of the features of a target optimisation problem. In contrast to most static (feature-independent) algorithm tuning engines such as irace and SPOT, our approach aims to derive the best parameter configuration of a given algorithm for a specific problem, exploiting the relationships between the algorithm parameters and the features of the problem. We formulate the problem of finding the relationships between the parameters and the problem features as a symbolic regression problem and we use genetic programming to extract these expressions in a human-readable form. For the evaluation, we apply our approach to the configuration of the (1 + 1) EA and RLS algorithms for the OneMax, LeadingOnes, BinValue and Jump optimisation problems, where the theoretically optimal algorithm parameters to the problems are available as functions of the features of the problems. Our study shows that the found relationships typically comply with known theoretical results – this demonstrates (1) the potential of model-based parameter tuning as an alternative to existing static algorithm tuning engines, and (2) its potential to discover relationships between algorithm performance and instance features in human-readable form.

KW - Genetic programming

KW - Model-based tuning

KW - Parameter tuning

UR - http://www.scopus.com/inward/record.url?scp=85107371096&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-72904-2_4

DO - 10.1007/978-3-030-72904-2_4

M3 - Article in proceeding

AN - SCOPUS:85107371096

SN - 9783030729035

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

SP - 51

EP - 67

BT - Evolutionary Computation in Combinatorial Optimization - 21st European Conference, EvoCOP 2021, Held as Part of EvoStar 2021, Proceedings

A2 - Zarges, Christine

A2 - Verel, Sébastien

PB - Springer Science+Business Media

Y2 - 7 April 2021 through 9 April 2021

ER -

El Yafrani M, Scoczynski M, Sung I, Wagner M, Doerr C, Nielsen P. MATE: A Model-Based Algorithm Tuning Engine: A Proof of Concept Towards Transparent Feature-Dependent Parameter Tuning Using Symbolic Regression. In Zarges C, Verel S, editors, Evolutionary Computation in Combinatorial Optimization - 21st European Conference, EvoCOP 2021, Held as Part of EvoStar 2021, Proceedings. Springer Science+Business Media. 2021. p. 51-67. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 12692 LNCS). doi: 10.1007/978-3-030-72904-2_4

MATE: A Model-Based Algorithm Tuning Engine: A Proof of Concept Towards Transparent Feature-Dependent Parameter Tuning Using Symbolic Regression

Abstract

Conference

Bibliographical note

Keywords

Access to Document

AUB Link

Other files and links

Fingerprint

Cite this