Broad Area Colloquium For AI-Geometry-Graphics-Robotics-Vision

Exponential Models in Language Modeling and Machine Learning

John Lafferty
School of Computer Science
Carnegie Mellon University

Monday, Mar 12, 2001, 4:15PM


Random fields and exponential models are well-studied and widely used methods in statistics, physics, computer vision, and other areas of computational science. In this talk we present new uses of and results related to this family of models, motivated from problems in natural language processing and machine learning. First, for segmenting and labeling sequences, we present a framework based on conditional random fields, which offers several advantages over hidden Markov models and stochastic grammars, the most commonly used tools for such tasks. Second, we derive an equivalence between the well-known technique of boosting in machine learning and maximum likelihood for exponential models. In both cases, the idea of using unnormalized models plays a key role.

About the Speaker

John Lafferty is an Associate Professor at Carnegie Mellon University, where he holds a joint appointment in the Computer Science Department, the Language Technologies Institute, and the Center for Automated Learning and Discovery. He has been on the CMU faculty since 1994. Prior to joining CMU, Dr. Lafferty was a Research Staff Member at the IBM Thomas J. Watson Research Center in Yorktown Heights, working on statistical methods for language processing. Dr. Lafferty received the Ph.D. in Mathematics from Princeton University, where he was a also member of the Program in Applied and Computational Mathematics. He is a past editorial board member of Computational Linguistics, and is currently an action editor for the Journal of Machine Learning Research, and the Machine Learning journal. His research interests include statistical learning algorithms, natural language processing, information retrieval, and coding and information theory.


Back to the Colloquium Page