Nuclear Discrepancy for Active Learning

Tom J. Viering, Jesse H. Krijthe, Marco Loog

Active learning algorithms propose which unlabeled objects should be queried for their labels to improve a predictive model the most. We study active learners that minimize general…