Sihong Xie     

Sihong Xie
Ph.D. Candidate
Department of Computer Science
University of Illinois at Chicago

"Decision Fusion via Consensus Maximization:
Synthesizing the power of multiple sources"

Thursday, February 18, 4:00 PM
STEPS room 101

Abstract:  Big data offer a great variety of data sources that are generating huge amount of information in an unprecedented speed. By synthesizing the power of multiple sources, with increased reliability, one can obtain more insightful knowledge that’s not available from any single source. As a special case, decision fusion is required by a wide range of real world applications, including but not limited to web data management, crowdsourcing, wireless sensor networks and healthcare. However, current fusion approaches are not geared to the full exploitation of the sources with different levels of accuracy and information completeness. The fusion of certain structured decisions, such as multi-labeled decisions, is also underdeveloped. In order to make methodological breakthroughs, an understanding of the theoretical limits of the current approaches is also in demand. In this talk, I will discuss my research work, including models and algorithms that help us better capture the high variety in the accuracy and the rich structures of the decisions from multiple sources. I will present two algorithms that can estimate source accuracy and the underlying structures of the decisions to improve the fusion. I will also describe how to formulate the fusion problem using structural risk minimization, a well-studied machine learning theory, to provide an unexplored perspective of the topic, along with a novel and effective fusion algorithm. I conclude the talk with a discussion of the keys to successful fusion models under various circumstances, with further insights into the developments of the topic in the future.

Bio:  Sihong Xie is a Ph.D. candidate at the Department of Computer Science in the University of Illinois at Chicago. His research is advised by Prof. Philip S. Yu and spans many facets of big data, including multiple source fusion, online content trustworthiness, social network mining and streaming learning algorithms. He has published 27 papers in the leading data mining conferences with over 360 citations. He holds M.Eng and BS in software engineering from Sun Yat-Sen University in China.

© 2014-2016 Computer Science and Engineering, P.C. Rossin College of Engineering & Applied Science, Lehigh University, Bethlehem PA 18015.