Active Learning of Binary Classifiers

Nina Balcan
Steve Hanneke

Abstract

In supervised machine learning, well-designed interactive learning algorithms can provide valuable improvements over passive algorithms in learning performance while reducing the amount of effort required of a human annotator. This has been observed in practice, but only recently have we begun to understand the conditions under which active learning can and cannot yield significant improvements. In this talk, we will survey some recent progress toward provably good active learning algorithms for learning linear separators. We will also briefly cover some recent progress toward understanding the sample complexity of active learning in general. Slides: Active Learning of Binary Classifiers

Back to the Main Page