Monday, October 24
12.10 - 1.00
CSE 4140

Speaker: Nikunj Oza, NASA Ames Research Center
Title: A Bi-Criterion Approach to Sparse One-class SVMs

Abstract:
In this talk, I will first give a brief overview of the Data Mining research done for NASA's Aviation Safety Program. I will then give a deeper discussion of one recent accomplishment---a variation of the One-class Support Vector Machines (SVMs) learning algorithm that produces sparser solutions faster compared to the classical algorithm. The proposed technique returns an approximate solution, nearly as good as the solution obtained by the classical approach, by minimizing the original risk function along with a regularization term. The outcome of the proposed learning technique was compared with the benchmark one-class SVM learning algorithm which often leads to solutions with redundant support vectors. We have tested the proposed algorithm on a variety of data sources under different conditions to demonstrate its effectiveness. In all cases the proposed algorithm closely preserves the accuracy of standard one-class SVMs while reducing both training time and test time by several factors.

Speaker Bio:
Nikunj C. Oza received his B.S. in Mathematics with Computer Science from the Massachusetts Institute of Technology (MIT) in 1994. He received his M.S. (in 1998) and Ph.D. (in 2001) in Computer Science from the University of California at Berkeley. He then joined NASA Ames Research Center and is a member of their Intelligent Data Understanding (IDU) group. He leads a team applying data mining to the problem of health management for aerospace vehicles. His 40+ research papers represent his research interests which include machine learning (especially ensemble learning and online learning), data mining, fault detection, integration of machine learning with automated planning and other areas of Artificial Intelligence, and their applications to Aeronautics and Earth Science. He received the 2007 Arch T. Colwell Award in 2007 for co-authoring one of the five most innovative technical papers selected out of over 3300 SAE technical papers published in 2005.