Uncertainty sampling and transductive experimental design for active dual supervision

摘要: Dual supervision refers to the general setting of learning from both labeled examples as well features. Labeled features are naturally available in tasks such text classification where it is frequently possible provide domain knowledge form words that associate strongly with a class. In this paper, we consider novel problem active dual supervision, or, how optimally query an example and feature labeling oracle simultaneously collect two different forms objective building best classifier most cost effective manner. We apply classical uncertainty experimental design based schemes graph/kernel-based models. Empirical studies confirm potential these significantly reduce acquiring data for training high-quality