Sequential Active Learning by Transductive Inference
Abstract: There has been recently a growing interest in the use of transductive inference for learning. We expand here the scope of transductive inference to active learning in a stream based setting. Towards that end, a novel active learning algorithm based on transduction (QBT) is proposed. QBT queries the label of an example based on the p-values obtained using transduction. We show that QBT is closely related to Query-by-Committee (QBC) using relations between transduction, Bayesian statistical testing, Kullback-Leibler divergence, and Shannon information. The feasibility and utility of QBT is shown on both binary and multi-class classification tasks using SVM as the choice classifier. Our experimental results show that QBT compares favorably, in terms of mean generalization, against random sampling, committee-based active learning, margin-based active learning, and QBC in the stream based setting.
Patent:
S.-S. Ho and H. Wechsler, System and Method for Active Learning/Modeling for Field Specifi.c Data Streams, US Patent Application No. 12/466,685, .filed in May 2009.
Code:
Tutorial:
“Conformal Predictions for Reliable Machine Learning: Theory and Applications”, organizers/presenters: Vineeth N Balasubramanian (Arizona State University), Shen-Shyang Ho, Sethuraman Panchanathan (Arizona State University), Vladimir Vovk (Royal Holloway, University of London), Tutorial Session, IJCNN 2011, San Jose, CA, 31 July 2011.
References:
S.-S. Ho and H. Wechsler, Query by Transduction, in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol .30, no. 9, 2008, pp 1557-1571.
S.-S. Ho, “Learning from Data Streams Using Transductive Inference and Martingale”, PhD Dissertation, Jan 2007.
S-S Ho and H. Wechsler, Learning from data streams via online transduction , ICDM 2004 Workshop on Temporal Data Mining: Algorithms, Theory and Applications (TDM 2004), Brighton, UK, Nov 2004.
S-S Ho and H. Wechsler, Transductive Confidence Machines for Active Learning, Proc. Int. Joint Conf. Neural Network (IJCNN 2003), Seattle, July 2003.