Exploring the Power of Source Reliability in Information Integration

Houping Xiao
Ph.D. Candidate, Department of Computer Science and Engineering
University at Buffalo, State University of New York
Location: 
SERC 306
Date: 
Wednesday, February 28, 2018 - 11:00
In this talk, I will present my work on the development of novel information integration methods that incorporate the estimation of source reliability. I explored the power of source reliability estimation in both data-level and model-level information integration. In data-level information integration, the objective is to jointly estimate which source is reliable and which piece of information is correct. To this end, I developed a new approach that iteratively calculates source reliability and true value estimates. In addition, the proposed approach outputs the confidence in the true value estimates, which provide useful information for decision making. In model-level information integration, I proposed to integrate models derived from a collection of distributed data sources. The proposed approach combines these models via weighted aggregation, in which weights reflect source reliability and are inferred jointly with the model combination procedure. An effective solution is designed through solving a constrained optimization problem by alternating direction method of multipliers. In both work, I proved some nice properties of the proposed approaches via theoretical analysis, and demonstrated their impact on some real applications such as indoor floorplan construction and crowdsourced question answering. I will also briefly discuss my other work on information integration, and their applications in health care, cyber security, and urban computing.