Tree Ensembles for Contextual Bandits
Journal article, 2024
We propose a new framework for contextual multi-armed bandits based on tree ensem-bles. Our framework adapts two widely used bandit methods, Upper Confidence Bound and Thompson Sampling, for both standard and combinatorial settings. As part of this frame-work, we propose a novel method of estimating the uncertainty in tree ensemble predictions. We further demonstrate the effectiveness of our framework via several experimental studies, employing XGBoost and random forests, two popular tree ensemble methods. Compared to state-of-the-art methods based on decision trees and neural networks, our methods ex-hibit superior performance in terms of both regret minimization and computational runtime, when applied to benchmark datasets and the real-world application of navigation over road networks.
Author
Hannes Nilsson
Chalmers, Computer Science and Engineering (Chalmers)
University of Gothenburg
Rikard Johansson
Chalmers, Computer Science and Engineering (Chalmers)
University of Gothenburg
Niklas Åkerblom
Chalmers, Computer Science and Engineering (Chalmers)
Volvo Cars
University of Gothenburg
Morteza Haghir Chehreghani
University of Gothenburg
Chalmers, Computer Science and Engineering (Chalmers)
Transactions on Machine Learning Research
28358856 (eISSN)
Vol. 2024Subject Categories (SSIF 2025)
Probability Theory and Statistics
Computer Sciences
DOI
10.48550/arXiv.2402.06963