A recent arXiv preprint tackles the problem of online linear classification when labels are expensive. In this setting, unlabeled samples arrive continuously, and each annotation carries a cost, so the learner must decide carefully which samples to label and at what point in the stream.

The paper frames this as a question of optimal query policy: a rule that prescribes whether and when a sample should be selected for annotation. The authors investigate how such selection schedules can be designed to make efficient use of the annotation budget.

Because only this single source was reviewed, the article does not compare the approach with other active-learning methods. The contribution is presented as a step toward understanding optimal selection schedules in online classification.