arxiveess.IVcs.LG2026-07-01
Group-invariant Coresets for Data-efficient Active Learning
L. C. Ayres, J. C. M. Bermudez, S. J. M. de Almeida, R. A. Borsoi
Active learning reduces labeling cost by querying the most informative unlabeled samples, but standard coreset methods ignore known data symmetries and can waste budget on transformed versions of the same instance. We propose GRINCO, a group-invariant coreset framework that perfo…