News Archives

RSS Feed

[Colloquium] Gaining Insight from Data by Learning Graphical Model Structures

November 6, 2014

Watch Colloquium:

MOV FILE

  • Date: Thursday, November 6, 2014
  • Time: 11:00 --- 12:15 PM
  • Place: Dane Smith Hall 125


Abstract

Machine learning algorithms for graphical model structure have helped scientists to gain insight into gene interaction networks, functional brain activity networks, information sharing networks, etc. Typical graph structure learning algorithms optimize for distribution matching; that is, they find the best graphical model to fit the empirical joint distribution of observed data. However, when answering practical questions; such as, how disease disrupts gene signaling networks; distribution matching may not lead to ideal solutions. In other applications, human experts are able to describe relationships among variables that cannot be inferred statistically and so distribution matching must be combined with this expert knowledge. I will present ongoing research projects in which we learn graphical models with objective functions modified to produce solutions that go beyond distribution matching to answer interesting questions about data.

Bio

Diane Oyen is a Postdoctoral Research Associate in the Space Data Systems group at Los Alamos National Laboratory. She received her B.S. in Electrical and Computer Engineering from Carnegie Mellon University in 1998 and her Ph.D. in Computer Science from the University of New Mexico in 2013. Her research interests include machine learning, interactive data exploration and knowledge discovery in data intensive scientific research.