Computer Science

Colloquium Details

Learning from Structured Data: Models and Applications

Abstract

Structured-input/output data emerge rapidly in a large number of applications, such as computational biology, social network analysis and climate data analysis. In this talk, I will examine two tasks under this topic: one is given the data with underlying structures, how we can learn the graph structures automatically. Specifically, we develop Granger temporal models, an emerging collection of graphical model techniques that allow us to model causal relationships from time series data by appealing Granger causality with success in biology and climate application; the other tasks is given the data with structured-input, how we can make use of the structure information for better modeling. Specifically, we develop Topic-Link LDA model, a Bayesian hierarchical model for topic modeling and social network analysis from linked blog data.

Biography

Yan Liu is a Research Staff Member at IBM TJ Watson Research. She received her M.Sc and Ph.D. degree in Language Technologies Institute, School of Computer Science from Carnegie Mellon University in 2004 and 2006. Her research interest includes machine learning and data mining algorithms for large-scale business analytics, social network study, computational biology and climate modeling. She has received several awards, including 2007 ACM Dissertation Award Honorable Mention, best application paper award in SDM 2007, winner of KDD Cup 2007, 2008 and INFORMS data mining competition 2008. She has published over 30 referred articles and served as a program committee of SIGKDD, CIKM, SIGIR, and several workshops in NIPS and ICDM.