Robust Entropy Search for Safe Efficient Bayesian Optimization
The practical use of Bayesian Optimization (BO) in engineering applications imposes special requirements: high sampling efficiency on the one hand and finding a robust solution on the other hand. We address the case of adversarial robustness, where all parameters are controllable during the optimization process, but a subset of them is uncontrollable or even adversely perturbed at the time of application. To this end, we develop an efficient information-based acquisition function that we call Robust Entropy Search (RES). We empirically demonstrate its benefits in experiments on synthetic and real-life data. The results show that RES reliably finds robust optima, outperforming state-of-the-art algorithms.
- Published in:
Proceedings of the Fortieth Conference on Uncertainty in Artificial Intelligence - Type:
Inproceedings - Authors:
Weichert, Dorina; Kister, Alexander; Houben, Sebastian; Link, Patrick; Ernis, Gunar - Year:
2024
Citation information
Weichert, Dorina; Kister, Alexander; Houben, Sebastian; Link, Patrick; Ernis, Gunar: Robust Entropy Search for Safe Efficient Bayesian Optimization, Proceedings of the Fortieth Conference on Uncertainty in Artificial Intelligence, 2024, 244, 3711--3729, PMLR, https://proceedings.mlr.press/v244/weichert24a.html, Weichert.etal.2024a,
@Inproceedings{Weichert.etal.2024a,
author={Weichert, Dorina; Kister, Alexander; Houben, Sebastian; Link, Patrick; Ernis, Gunar},
title={Robust Entropy Search for Safe Efficient Bayesian Optimization},
booktitle={Proceedings of the Fortieth Conference on Uncertainty in Artificial Intelligence},
volume={244},
pages={3711--3729},
publisher={PMLR},
url={https://proceedings.mlr.press/v244/weichert24a.html},
year={2024},
abstract={The practical use of Bayesian Optimization (BO) in engineering applications imposes special requirements: high sampling efficiency on the one hand and finding a robust solution on the other hand. We address the case of adversarial robustness, where all parameters are controllable during the optimization process, but a subset of them is uncontrollable or even adversely perturbed at the time of...}}
