Safe Bayesian Optimization with Counterfactual Policies

2026-07-06

AI Summary

This work addresses the challenge of ensuring that newly deployed policies in decision optimization do not underperform relative to a baseline, despite the unobservability of counterfactual outcomes under the baseline policy, which complicates reliable enforcement of safety constraints. To overcome this, the study introduces conformal prediction into counterfactual inference, constructing valid uncertainty intervals for baseline outcomes. These intervals are integrated with Bayesian optimization to adaptively calibrate safety constraints under covariate shift. The proposed method rigorously controls the constraint violation rate at a user-specified confidence level, offering theoretical guarantees of safety. Empirical validation and sensitivity analyses further demonstrate its practical effectiveness and robustness.

Abstract

In many decision-making settings, new interventions are acceptable only if they do not reduce outcomes below some established threshold. For example, in clinical medicine, new treatments are often acceptable only if they do not worsen outcomes relative to an established standard of care. Safe Bayesian optimization maximizes an objective subject to safety constraints. In the setting that we consider here, safety is defined relative to a known baseline policy whose outcomes are counterfactual and therefore unobserved. Thus, the counterfactual outcomes of the baseline policy must be estimated and those (uncertain) estimates must be used to safely optimize the objective. We address this estimation problem by using conformal prediction to construct valid uncertainty intervals for counterfactual baseline outcomes, and we show how these intervals can be integrated into safe Bayesian optimization to ensure that constraint violations occur at or below a user-specified rate. We also show how to adapt these conformal estimates to different kinds of covariate shift. We provide a safety proof, experimental evidence, and a sensitivity analysis.

Problem

• Safe Bayesian Optimization
• Counterfactual Policy
• Safety Constraints
• Conformal Prediction
• Covariate Shift

Innovation

• Safe Bayesian Optimization
• Counterfactual Policy
• Conformal Prediction
• Uncertainty Quantification
• Covariate Shift