MO-SMAC: Multi-objective Sequential Model-based Algorithm Configuration

verfasst von

Jeroen Rook, Carolin Benjamins, Jakob Bossek, Heike Trautmann, Holger Hoos, Marius Lindauer

Abstract

Automated algorithm configuration aims at finding well-performing parameter configurations for a given problem, and it has proven to be effective within many AI domains, including evolutionary computation. Initially, the focus was on excelling in one performance objective, but, in reality, most tasks have a variety of (conflicting) objectives. The surging demand for trustworthy and resource-efficient AI systems makes this multi-objective perspective even more prevalent. We propose a new general-purpose multi-objective automated algorithm configurator by extending the widely-used SMAC framework. Instead of finding a single configuration, we search for a non-dominated set that approximates the actual Pareto set. We propose a pure multi-objective Bayesian Optimisation approach for obtaining promising configurations by using the predicted hypervolume improvement as acquisition function. We also present a novel intensification procedure to efficiently handle the selection of configurations in a multi-objective context. Our approach is empirically validated and compared across various configuration scenarios in four AI domains, demonstrating superiority over baseline methods, competitiveness with MO-ParamILS on individual scenarios and an overall best performance.

Organisationseinheit(en)

Fachgebiet Maschinelles Lernen
Institut für Künstliche Intelligenz
Forschungszentrum L3S

Externe Organisation(en)

University of Twente
Universität Paderborn
Leiden University
Rheinisch-Westfälische Technische Hochschule Aachen (RWTH)
University of British Columbia

Typ

Artikel

Journal

Evolutionary computation

Band

25

Seiten

1-25

ISSN

1063-6560

Publikationsdatum

10.03.2025

Publikationsstatus

Veröffentlicht

Peer-reviewed

Ja

Elektronische Version(en)

https://doi.org/10.1162/evco_a_00371 (Zugang: Geschlossen)