Ensemble Robust Bayesian Optimisation Algorithm

A new algorithm for Ensemble Distributionally Robust Bayesian Optimisation (EDRBO) has been proposed, addressing zeroth-order optimisation under context distributional uncertainty. The method uses an ensemble as a surrogate model to enhance robustness against noisy data, a common challenge in Bayesian optimisation (BO). It remains computationally tractable while handling continuous contexts. Theoretical sublinear regret bounds are established, improving upon current state-of-the-art results. Empirical evaluations confirm alignment with theoretical guarantees. The work is published on arXiv under computer science and machine learning categories.