Multi-armed bandits for trustworthy and resource-efficient algorithm configuration / Jasmin Brandt ; 1st Reviewer: Prof. Dr. Eyke Hüllermeier (Institute of Informatics Ludwig Maximilian University of Munich), 2nd Reviewer: Prof. Dr. Kevin Tierney (Decision and Operation Technologies Group Bielefeld University), Supervisor: Prof. Dr. Eyke Hüllermeier. Paderborn, 2025
Content
- Titlepage
- Abstract
- Zusammenfassung
- Acknowledgement
- Contents
- 1 Introduction
- 2 Algorithm Configuration
- 3 Multi-Armed Bandits
- 4 State-Of-The-Art and Contributions
- 4.1 Finding Optimal Arms in Non-stochastic Combinatorial Bandits
- 4.2 AC-Band
- 4.3 Incremental Successive Halving and Incremental Hyperband
- 5 Finding Optimal Arms in Non-stochastic Combinatorial Bandits with Semi-bandit Feedback and Finite Budget
- 6 AC-Band: A Combinatorial Bandit-Based Approach to Algorithm Configuration
- 7 Best Arm Identification with Retroactively Increased Sampling Budget for More Resource-Efficient HPO
- 8 Conclusion and Outlook
- Bibliography
- A Appendix to Finding Optimal Arms in Non-stochastic Combinatorial Bandits with Semi-bandit Feedback and Finite Budget
- B Appendix to AC-Band: A Combinatorial Bandit-Based Approach to Algorithm Configuration
- C Appendix to Best Arm Identification with Retroactively Increased Sampling Budget for More Resource-Efficient HPO
- List of Figures
- List of Symbols
- Colophon
- Declaration