Seminar: Stream Data Mining: Active Labeling, Cleansing, and Vague Learning

Presenter: A/Prof Xingquan (Hill) Zhu, QCIS, FEIT, University of Technology, Sydney

Abstract: In this talk, I will summarize a number of steam data mining problems: Active Labeling, Cleansing, and Vague Learning, that we have addressed in recent years. Regarding active labeling, we consider that labeling all stream data is expensive and impractical, so our objective is to label a small portion of stream data from which a model is derived to predict newly arrived instances as accurately as possible. For incorrectly labeled training samples of data streams, we propose a Maximum Variance Margin principle to accurately identify and remove mislabeled data, so that the prediction models built from the cleansed streams will be more accurate than those trained from the raw, noisy streams. For vague learning in data streams, we allow users to label instance groups, instead of single  instances, as positive samples for learning. Experimental results on synthetic and real-world data demonstrate the superior performances of our research, in comparison with other simple approaches.

Brief Bio: Xingquan Zhu received his PhD degree in Computer Science from Fudan University, Shanghai China, in 2001. He is an Associate Professor of the Faculty of Engineering and Information Technology, University of Technology, Sydney (UTS), Australia. His research focuses on data mining, machine learning, and multimedia systems. Since 2000, he has published more than 100 referred journal and conference proceedings papers in these areas (12 A*/A journal articles and 35 A conference proceeding papers ranked by ARC ERA). Dr. Zhu served or is serving on the editorial board (Associate Editor) for seven international journals (including one IEEE Transactions). He is the program committee co-chair for the 9th IEEE International Conference on Machine Learning and Applications (ICMLA-10). He founded the International Workshop on Mining Multiple information sources in 2007, and has successfully organized the workshop for three times (in conjunction with KDD-07, KDD-08, and ICDM-09). In addition, He also served or is currently serving on the program committee for more than 70 times for international conferences and symposiums.