Synergies of Deep and Classical Exploratory Landscape Features for Automated Algorithm Selection

Moritz Seiler; Urban Škvorc; Carola Doerr; Heike Trautmann

doi:10.1007/978-3-031-75623-8_29

Synergies of Deep and Classical Exploratory Landscape Features for Automated Algorithm Selection

Moritz Seiler^*, Urban Škvorc, Carola Doerr, Heike Trautmann

^*Corresponding author for this work

Datamanagement & Biometrics

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

2 Citations (Scopus)

Abstract

Per-instance automated algorithm selection (AAS) aims at leveraging the complementarity of optimization algorithms with respect to different problem types. State-of-the-art AAS methods for numerical black-box optimization rely on supervised learning techniques that are supported by exploratory landscape analysis (ELA) feature sets. Recent works question the generalization ability of popular AAS approaches, which motivated the design of alternative feature sets. In this work, we take a closer look at the recently proposed set of Deep ELA features and investigate the ways in which Deep ELA complements the classical ELA feature sets. To this end, we first study the correlation between the two feature collections, both through pairwise classification and through regression models. The complementarity observed in these analyses is confirmed by an AAS study, where models combining deep and classical features outperform those that are restricted to selecting from only of the two collections.

Original language	English
Title of host publication	Learning and Intelligent Optimization
Subtitle of host publication	18th International Conference, LION 18, Revised Selected Papers
Editors	Paola Festa, Daniele Ferone, Tommaso Pastore, Ornella Pisacane
Publisher	Springer
Pages	361-376
Number of pages	16
ISBN (Electronic)	978-3-031-75623-8
ISBN (Print)	978-3-031-75622-1
DOIs	https://doi.org/10.1007/978-3-031-75623-8_29
Publication status	Published - 2025
Event	18th International Conference on Learning and Intelligent Optimization, LION 2024 - Ischia Island, Italy Duration: 9 Jun 2024 → 13 Jun 2024 Conference number: 18

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	14990 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	18th International Conference on Learning and Intelligent Optimization, LION 2024
Abbreviated title	LION 2024
Country/Territory	Italy
City	Ischia Island
Period	9/06/24 → 13/06/24

Keywords

n/a OA procedure
Black-box Optimization
Deep Learning
Exploratory Landscape Analysis
Feature Selection
Self-Supervised Learning
Automated Algorithm Selection

Access to Document

10.1007/978-3-031-75623-8_29

Cite this

Seiler, M., Škvorc, U., Doerr, C., & Trautmann, H. (2025). Synergies of Deep and Classical Exploratory Landscape Features for Automated Algorithm Selection. In P. Festa, D. Ferone, T. Pastore, & O. Pisacane (Eds.), Learning and Intelligent Optimization: 18th International Conference, LION 18, Revised Selected Papers (pp. 361-376). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14990 LNCS). Springer. https://doi.org/10.1007/978-3-031-75623-8_29

Seiler, Moritz ; Škvorc, Urban ; Doerr, Carola et al. / Synergies of Deep and Classical Exploratory Landscape Features for Automated Algorithm Selection. Learning and Intelligent Optimization: 18th International Conference, LION 18, Revised Selected Papers. editor / Paola Festa ; Daniele Ferone ; Tommaso Pastore ; Ornella Pisacane. Springer, 2025. pp. 361-376 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{b7a5de403f3c4319a795f103fced87c9,

title = "Synergies of Deep and Classical Exploratory Landscape Features for Automated Algorithm Selection",

abstract = "Per-instance automated algorithm selection (AAS) aims at leveraging the complementarity of optimization algorithms with respect to different problem types. State-of-the-art AAS methods for numerical black-box optimization rely on supervised learning techniques that are supported by exploratory landscape analysis (ELA) feature sets. Recent works question the generalization ability of popular AAS approaches, which motivated the design of alternative feature sets. In this work, we take a closer look at the recently proposed set of Deep ELA features and investigate the ways in which Deep ELA complements the classical ELA feature sets. To this end, we first study the correlation between the two feature collections, both through pairwise classification and through regression models. The complementarity observed in these analyses is confirmed by an AAS study, where models combining deep and classical features outperform those that are restricted to selecting from only of the two collections.",

keywords = "n/a OA procedure, Black-box Optimization, Deep Learning, Exploratory Landscape Analysis, Feature Selection, Self-Supervised Learning, Automated Algorithm Selection",

author = "Moritz Seiler and Urban {\v S}kvorc and Carola Doerr and Heike Trautmann",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.; 18th International Conference on Learning and Intelligent Optimization, LION 2024, LION 2024 ; Conference date: 09-06-2024 Through 13-06-2024",

year = "2025",

doi = "10.1007/978-3-031-75623-8\_29",

language = "English",

isbn = "978-3-031-75622-1",

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

publisher = "Springer",

pages = "361--376",

editor = "Paola Festa and Daniele Ferone and Tommaso Pastore and Ornella Pisacane",

booktitle = "Learning and Intelligent Optimization",

address = "Germany",

}

Seiler, M, Škvorc, U, Doerr, C & Trautmann, H 2025, Synergies of Deep and Classical Exploratory Landscape Features for Automated Algorithm Selection. in P Festa, D Ferone, T Pastore & O Pisacane (eds), Learning and Intelligent Optimization: 18th International Conference, LION 18, Revised Selected Papers. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 14990 LNCS, Springer, pp. 361-376, 18th International Conference on Learning and Intelligent Optimization, LION 2024, Ischia Island, Italy, 9/06/24. https://doi.org/10.1007/978-3-031-75623-8_29

Synergies of Deep and Classical Exploratory Landscape Features for Automated Algorithm Selection. / Seiler, Moritz; Škvorc, Urban; Doerr, Carola et al.
Learning and Intelligent Optimization: 18th International Conference, LION 18, Revised Selected Papers. ed. / Paola Festa; Daniele Ferone; Tommaso Pastore; Ornella Pisacane. Springer, 2025. p. 361-376 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14990 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - Synergies of Deep and Classical Exploratory Landscape Features for Automated Algorithm Selection

AU - Seiler, Moritz

AU - Škvorc, Urban

AU - Doerr, Carola

AU - Trautmann, Heike

N1 - Conference code: 18

PY - 2025

Y1 - 2025

N2 - Per-instance automated algorithm selection (AAS) aims at leveraging the complementarity of optimization algorithms with respect to different problem types. State-of-the-art AAS methods for numerical black-box optimization rely on supervised learning techniques that are supported by exploratory landscape analysis (ELA) feature sets. Recent works question the generalization ability of popular AAS approaches, which motivated the design of alternative feature sets. In this work, we take a closer look at the recently proposed set of Deep ELA features and investigate the ways in which Deep ELA complements the classical ELA feature sets. To this end, we first study the correlation between the two feature collections, both through pairwise classification and through regression models. The complementarity observed in these analyses is confirmed by an AAS study, where models combining deep and classical features outperform those that are restricted to selecting from only of the two collections.

AB - Per-instance automated algorithm selection (AAS) aims at leveraging the complementarity of optimization algorithms with respect to different problem types. State-of-the-art AAS methods for numerical black-box optimization rely on supervised learning techniques that are supported by exploratory landscape analysis (ELA) feature sets. Recent works question the generalization ability of popular AAS approaches, which motivated the design of alternative feature sets. In this work, we take a closer look at the recently proposed set of Deep ELA features and investigate the ways in which Deep ELA complements the classical ELA feature sets. To this end, we first study the correlation between the two feature collections, both through pairwise classification and through regression models. The complementarity observed in these analyses is confirmed by an AAS study, where models combining deep and classical features outperform those that are restricted to selecting from only of the two collections.

KW - n/a OA procedure

KW - Black-box Optimization

KW - Deep Learning

KW - Exploratory Landscape Analysis

KW - Feature Selection

KW - Self-Supervised Learning

KW - Automated Algorithm Selection

UR - https://www.scopus.com/pages/publications/85216119477

U2 - 10.1007/978-3-031-75623-8_29

DO - 10.1007/978-3-031-75623-8_29

M3 - Conference contribution

AN - SCOPUS:85216119477

SN - 978-3-031-75622-1

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

SP - 361

EP - 376

BT - Learning and Intelligent Optimization

A2 - Festa, Paola

A2 - Ferone, Daniele

A2 - Pastore, Tommaso

A2 - Pisacane, Ornella

PB - Springer

T2 - 18th International Conference on Learning and Intelligent Optimization, LION 2024

Y2 - 9 June 2024 through 13 June 2024

ER -

Seiler M, Škvorc U, Doerr C, Trautmann H. Synergies of Deep and Classical Exploratory Landscape Features for Automated Algorithm Selection. In Festa P, Ferone D, Pastore T, Pisacane O, editors, Learning and Intelligent Optimization: 18th International Conference, LION 18, Revised Selected Papers. Springer. 2025. p. 361-376. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). Epub 2025 Jan 3. doi: 10.1007/978-3-031-75623-8_29