Goal-driven Bayesian Optimal Experimental Design for Robust Decision-Making Under Model Uncertainty

2026-05-25

AI Summary

Traditional Bayesian optimal experimental design (BOED) aims to maximize information gain about model parameters but does not necessarily improve the quality of downstream decisions. This work proposes GoBOED, a novel framework that explicitly aligns experimental design with a specific decision objective. By jointly modeling an amortized variational posterior and a differentiable convex decision layer, GoBOED enables gradient-based, decision-oriented optimization. Theoretical analysis reveals that its gradients are insensitive to parameter directions irrelevant to the decision task, thereby uncovering a broader set of near-optimal designs. Empirical results in source localization, epidemic control, and pharmacokinetic tasks demonstrate that GoBOED produces designs better aligned with the intended decision goals and yields a significantly larger set of near-optimal solutions compared to conventional BOED approaches.

Abstract

Bayesian optimal experimental design (BOED) selects experiments to maximize information gain about model parameters. However, in decision-critical settings, reducing parameter uncertainty does not necessarily improve downstream decisions, as only specific parameter directions relevant to the objective truly matter. We propose GoBOED, a goal-driven BOED framework that directly optimizes experimental designs for a specified decision-making objective. GoBOED combines an amortized variational posterior surrogate with a differentiable convex decision layer, enabling gradient-based design optimization that is fully decision-focused. We theoretically show that GoBOED gradients are insensitive to parameter directions irrelevant to the decision objective, providing a formal justification for why goal-driven design achieves equivalent decision quality over a wider set of experimental designs than information-gain maximization. Empirically, across source localization, epidemic management, and pharmacokinetic control, GoBOED identifies designs that better align with downstream decision objectives and reveals that near-optimal design windows are substantially wider than those predicted by goal-agnostic BOED approaches.

Problem

• Bayesian optimal experimental design
• model uncertainty
• decision-making
• goal-driven
• information gain

Innovation

• goal-driven Bayesian experimental design
• decision-focused optimization
• amortized variational inference
• differentiable decision layer
• model uncertainty