Mathias Niepert     
Dr. Mathias Niepert

Postdoctoral Research Associate
University of Washington
"Tractable Probabilistic Models for Big Relational Data”
Monday, February 9, 4:00 PM
Packard Lab Room 466

Abstract: The extraction of knowledge from the world wide web and other big data sources is a central problem in the data sciences with numerous applications and wide-ranging implications for future technologies. The majority of the world's knowledge is expressed in natural language text and as such needs to be brought into structured form to be accessible for automated reasoning and analysis. There are several existing information extraction projects such as the Google Knowledge Graph, DBpedia, and NELL. These projects populate large relational knowledge bases from web text. Each of these projects, however, suffers from several shortcomings such as lack of coverage, error-prone and uncertain extractions, or rigid ontologies. Still,these knowledge bases are a valuable and ever-growing data set for bootstrapping statistical relational models of the world's knowledge. These models could enable innovative technologies such as search engines augmenting keyword-based results with entities, their properties and relations.

Since inference and learning in probabilistic models is NP-hard in general, we have to develop new theoretical foundations and algorithms that scale statistical relational models to large data sets. To this end, we present several lines of recent work. First, we present symmetry-aware inference and learning where we exploit both symmetries and conditional independence to design expressive yet tractable probabilistic models. Second, we present tractable probabilistic knowledge bases (TPKBs) featuring sublinear, disk-based, and parallel inference algorithms. We learn a TPKB from existing information extraction projects and apply it to data integration problems.

Bio:   Mathias Niepert is a postdoctoral research associate at the University of Washington in Seattle, working mainly with Pedro Domingos. He has a PhD from Indiana University and was a member of the Data and Web Science Research Group at the University of Mannheim, Germany. His work has won best paper awards at international conferences such as UAI, AAAI, IJCNLP, and ESWC. He is the principal investigator of a Google faculty research award and a bilateral DFG-NEH research award. Mathias is also co-founder of several open-source digital humanities projects such as the Indiana Philosophy Ontology Project and the Linked Humanities Project.

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