The abstract is read at the publisher; the summary is JClub's.
Cluster-based training set selection consistently outperforms active learning across 60 chemical and material property prediction tasks, with a theoretical Gaussian process variance basis in the low-data limit.
- Why it matters: When data are scarce, it is unclear which acquisition strategies are most efficient across many property prediction tasks, and what explains their relative performance.
- What they did: The authors surveyed 100 space-filling and active learning algorithms over 60 property prediction tasks, comparing selection from a predefined candidate library across representation and training set size settings.
- The result: A Gaussian process variance framework identifies cluster-based selection as the principled optimum in the low-data limit, supporting a broadly effective, data-efficient strategy for materials design.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.