Paper Overview
- Field: Machine Learning
- Authors: Xinyang Li, Kevin Stone, Ajit Vikram
- Published: 2026-09-21
- arXiv: 2609.24954
- Combines an optimistic joint feasibility mask with stochastic crossing extensions for multi-objective settings, concentrating sampling on the joint failure edge.
- Benchmark studies show JAREX recovers the joint pass region more accurately and with greater sample efficiency than factorial DOE, space-filling designs, and greedy per-objective strategies, across the full experimental budget.
- For batched experiments, JAREX reduces the number of iterative process characterization campaigns by more than half while maintaining high accuracy on boundary identification tasks.
Summary
Pharmaceutical process characterization is central to Quality by Design (QbD) because it defines how variations in process parameters affect the ability to meet product quality specifications, thereby supporting proven acceptable ranges and robust manufacturing. In practice, however, characterization still relies largely on factorial design of experiments (DOE) approaches, which are inefficient for resolving multivariate pass/fail boundaries in higher-dimensional spaces.
While Bayesian optimization has transformed process optimization, adaptive methods for multi-objective process characterization remain lacking. The authors introduce JAREX (Joint Acceptable Region EXploration), a Bayesian active-learning acquisition function for multi-objective process characterization. JAREX formulates characterization as a joint boundary-learning problem, adaptively selecting experiments to recover the joint pass region defined by multiple objectives simultaneously meeting threshold criteria.
Key Contributions
Availability
JAREX is implemented in the open-source obsidian package, providing a modular framework for adaptive, data-efficient, multi-objective algorithmic process characterization.
--- *Auto-collected on 2026-09-23.*