Abstract: Many applications are driven by evolving data. For instance, patterns in web traffic, program execution traces, network event logs, etc., are usually non-stationary. Data with evolving characteristics creates a life cycle for classifiers. Creating, updating, and recycling classifiers intelligently is important to the performance of predictive decision making. Furthermore, when resources are limited, it is a challenge to maintain good predictive power when only part of the data can be observed by the classifier. In this talk, I focus on two issues. One is how to break away from the state-of-the-art "chasing trends" approach, where learning is conducted continuously but patterns we capture is always one step behind the current trend. The other is how to decide when to spend and when to spare, with regard to limited bandwidth/CPU time in classification, for high quality decision making. I will also discuss applications that can benefit from the work. Encouraging early results scatter in many fields, including data mining, system management, network analysis, software engineering, etc., and even more challenges remain.

Short Bio: Haixun Wang is currently a research staff member at IBM T. J. Watson Research Center. He has been a Technical Assistant to Stuart Feldman, Vice President of Computer Science of IBM Research, since 2006. He received the B.S. and the M.S. degree, both in computer science, from Shanghai Jiao Tong University in 1994 and 1996. He received the Ph.D. degree in computer science from the University of California, Los Angeles in 2000. His main research interest is database language and systems, data mining, and information retrieval. He has published more than 100 research papers in referred international journals and conference proceedings. His recent professional activities include demo program chair of ICDE'09, senior PC member of KDD'08, and he served in program committees of various international conferences and workshops, and has been a reviewer for some leading academic journals in the database field.