Constrained Bayesian Experimental Design via Online Planning
It addresses the problem of adapting experimental designs to real-world constraints (e.g., budget, physical limits) for practitioners using Bayesian experimental design.
This work introduces a method for Bayesian experimental design that handles dynamic constraints by combining offline pre-training with online multi-step lookahead planning, achieving substantially more informative design sequences than existing methods across constrained tasks.
Bayesian experimental design (BED) is a principled framework for data-efficient design of sequential experiments. However, existing BED methods are unable to adapt to dynamic constraints inherent in real-world tasks due to budget limitations, varying costs, or physical constraints that restrict how designs evolve over time. In this paper, we introduce a novel approach to BED that enables constrained optimization of experimental designs by combining offline pre-training of an amortized policy and a posterior network with online multi-step lookahead planning using scenario trees. We empirically demonstrate that our method yields substantially more informative design sequences than existing methods across a range of constrained BED tasks, while incurring only a modest additional computational overhead.