A Survey of Methods for Automated Algorithm Configuration

Date: Wednesday, Sep 13, 2023, 14:30 - 16:00

  • Marcel Wever, MCML, LMU Munich
  • Jasmin Brandt, Paderborn University
  • Viktor Bengs, LMU Munich



Given a parametrized algorithm, a set of training problem instances, and one or multiple objective functions, algorithm configuration (AC) is concerned with the automated search for a suitable parameterization of the given algorithm. In this tutorial, we give an introduction to algorithm configuration in general and discuss different variants of this problem. Furthermore, the literature is rich of methods for algorithm configuration but many of them are tailored to the characteristics of specific problem variants. Therefore, we also provide a classification scheme for both algorithm configuration problems and methods for tackling them to distinguish the peculiarities of different methods. According to this classification scheme, we introduce various methods and discuss their advantages and disadvantages. Ultimately, we discuss open questions and challenges which are yet to be tackled.


Marcel Wever

Marcel Wever is a postdoctoral researcher at the chair of Artificial Intelligence and Machine Learning and Coordinator Education at the Munich Center for Machine Learning at LMU Munich. After his B.Sc. and M.Sc. in computer science at Paderborn University in 2015 and 2017 respectively, during his PhD, which he obtained in 2021 at Paderborn University, Marcel Wever worked on automated machine learning, algorithm selection, and algorithm configuration. Moreover, he regularly reviews for top-tier conferences as NeurIPS, ICML, AAAI, and ICLR and has been honoured as an outstanding reviewer at NeurIPS 2022, ICML 2021, and ICML 2020. At LMU Munich, Marcel Wever gives a Master course entitled ”Automated Algorithm Configuration and Design”. Furthermore, he gave a practical course and co-organized a Bachelor seminar on automated machine learning and algorithm selection respectively.

Viktor Bengs

Viktor Bengs obtained his Ph.D. in 2018 in Statistics at the Philipps-Universität Marburg, Germany. He is currently a postdoctoral researcher at the Chair of Artificial Intelligence and Machine Learning at the Institute of Informatics, LMU Munich, Germany. In summer 2022, he was an interim professor at the Chair of Statistical Learning and Data Science at the Institute of Statistics, LMU Munich. His research focuses on the development of theoretically sound algorithms for sequential decision tasks with weakly supervised feedback, e.g., preference information or censored observations, with automated algorithm configuration being the main application area. Despite his early career level, he has already published a couple of articles in top-tier journals and conferences and has also served the research community as a reviewer. His engagement in reviewing has been honored by being selected as a top reviewer at UAI 2021, AISTATS 2022 and UAI 2022 as well as an invitation to the editorial board of the Machine Learning journal. He will (co-)organize the workshop entitled “The Many Facets of Preference-Based Learning” at the upcoming ICML 2023 and give the first part of the tutorial about “Aleatoric and Epistemic Uncertainty in Statistics and Machine Learning” within the workshop “Uncertainty meets explainability in machine learning” at the upcoming ECML-PKDD 2023.

Jasmin Brandt

Jasmin Brandt is a PhD student in the Data Science Group at Paderborn University. She finished her B.Sc. in mathematics in 2017 and her M.Sc. in computer science in 2019, both at the University of Bonn. Her main research topics during her PhD are Multi-Armed Bandit Algorithms and their application to Algorithm Configuration, in which she already published papers in top-tier conferences like NeurIPS 2022 and AAAI 2023. Furthermore, Jasmin Brandt held the exercise group for the lecture ”foundations of intelligent systems”.