KolTa active learning framework discovers top formulations with limited queries
Original title: KolTa: A Physics-Informed Active Learning Framework for Formulation Discovery with Limited Data
KolTa selects batches for top 5% formulations using PRC-UCB, outperforming comparators on viscosity, solubility, and conductivity tasks.
KolTa is an active learning method combining three-stage predictor and physics constraints for formulation discovery. It uses PRC-UCB acquisition to select batches under fixed budget, targeting top 5% at given temperature. Evaluated on four tasks, it wins most paired comparisons and reduces queries needed by 18%.
- Combines component contributions, interactions, and temperature response
- PRC-UCB uses predicted utility, disagreement, and physics residual
- First or second in AULC on all tasks
- 18% fewer queries for 80% recovery on Drug task
- Adapts via component roles and physical constraints
From the original article
1 min read at arxiv.orgFinding high-performing formulations is difficult when few have been measured and experimental budgets are limited. We propose KolTa, an active learning framework that selects batches to discover the top 5% of formulations at a target temperature under a fixed query budget.
By Boyong Tang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.