Robust Entropy Search for Safe Efficient Bayesian Optimization

Dorina Weichert, Alexander Kister, Sebastian Houben, Patrick Link, Gunar Ernis

Keywords: Bayesian Optimization, Robustness, Information-Based Acquisition Functions

TL;DR: Derivation of an information-based adversarially robust acquisition function with superior performance on real-life and synthetic benchmarks.

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 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.

List Of Authors: Weichert, Dorina and Kister, Alexander and Houben, Sebastian and Link, Patrick and Ernis, Gunar

Code Url: https://github.com/fraunhofer-iais/Robust-Entropy-Search

Submission Number: 178