Ensemble Distributionally Robust Bayesian Optimisation

2026-05-08

AI Summary

This work addresses the robustness challenges in zeroth-order optimization arising from distributional uncertainty in continuous contextual settings by proposing a novel ensemble distributionally robust Bayesian optimization algorithm. The method uniquely integrates ensemble surrogate models with distributionally robust optimization, effectively mitigating the impact of distributional perturbations while preserving computational tractability. Theoretical analysis establishes that the proposed algorithm achieves a sublinear regret bound, providing strong convergence guarantees. Empirical evaluations demonstrate its superior performance over existing approaches, with experimental results aligning closely with the theoretical assurances.

Abstract

We study zeroth-order optimisation under context distributional uncertainty, a setting commonly tackled using Bayesian optimisation (BO). A prevailing strategy to make BO more robust to the complex and noisy nature of data is to employ an ensemble as the surrogate model, thereby mitigating the weaknesses of any single model. In this study, we propose a novel algorithm for Ensemble Distributionally Robust Bayesian Optimisation that remains computationally tractable while managing continuous context. We obtain theoretical sublinear regret bounds, improving current state-of-the-art results. We show that our method's empirical behaviour aligns with its theoretical guarantees.

Problem

• Bayesian optimisation
• distributional uncertainty
• zeroth-order optimisation
• ensemble methods
• contextual optimization

Innovation

• Ensemble
• Distributionally Robust
• Bayesian Optimisation
• Sublinear Regret
• Zeroth-order Optimisation