Title: Penalty based methods for multi-view and multi-task learning problems with network structure Speaker: George Michailidis, University of Michigan Date/Time: Wednesday, July 29, 2009, 1:00 – 2:00 pm Location: CSRI Building, Room 90 (Sandia NM) Brief Abstract: In many learning problems, the data exhibit a network (graph) structure either on the available attributes or on the observations. Further, the attributes can be naturally partitioned into subsets (views) and the same can occur on the observations (tasks). In this talk, we provide an overview of methods based on a regularization framework that uses penalty functions for such learning problems. The problems are motivated by a number of examples from the biological, engineering and social sciences and the methods illustrated using real and CSRI POC: Scott Mitchell, (505) 845-7594 |