Decision-Aware Information Acquisition for Optimization under Uncertainty

Information acquisition is often guided by how much uncertainty it can reduce about an underlying model. When the goal is optimization, however, the value of information lies in its impact on the final decision. We propose decision-aware information acquisition, a framework that selects information according to its expected impact on the downstream optimization decision, propagating posterior uncertainty through the optimization problem and prioritizing information that can reduce decision uncertainty. We develop a Bayesian instantiation with a decision-based preference model and test on two synthetic optimization problems.