AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-19
Active Learning for Interatomic Potentials: A Review of Uncertainty-Driven Data Efficiency
This review surveys more than seventy studies on active learning for machine-learning interatomic potentials, which aims to make potentials accurate for the conditions where they are applied. It covers on-the-fly molecular dynamics, fine-tuning of pretrained models, and sampling for materials and chemical systems. The authors also note that uncertainty estimates differ across architectures and are often least reliable where they matter most.