A Bayesian Active-Learning Acquisition Function Halves the Experiments Needed to Map a Multi-Objective Pass Region

arXiv·low signalXBlueskyLinkedInCopy link

JAREX (arXiv 2609.24954, 21 Sep 2026) applies Bayesian active learning to pharmaceutical process characterization, which still relies mostly on factorial design of experiments and is inefficient at resolving multivariate pass/fail boundaries in higher dimensions. It formulates characterization as joint boundary learning, adaptively selecting experiments to recover the region where threshold criteria across multiple objectives hold simultaneously, combining an optimistic joint-feasibility mask with a multi-objective extension of randomized straddle to focus sampling on the joint edge of failure. Benchmarks report more accurate and sample-efficient recovery than factorial DOE, space-filling designs and greedy objective-wise strategies, cutting batched iterative experiments by more than half at comparable accuracy.