Linear discriminant analysis (LDA) is widely used to reduce dimensionality in classification.  However, it suffers from the small sample size problem, where the number of examples is less than the number of dimensions or attributes.  In this talk, we propose to address the small sample size problem in the framework of statistical learning theory. We compute linear discriminants by regression, where the singularity problem is resolved. The resulting discriminants are complete in that they include both "regular" and "irregular" information.  We call our approach "Discriminant Learning Analysis." In addition, our approach allows us to establish an error bound for LDA. Finally we show experimental results validating our theoretical analysis.

Jing Peng is an Associate Professor of Computer Science at Montclair State University.  He received the Ph.D. degree in computer science from Northeastern University 1994. Previously, he was on the faculty of Electrical Engineering and Computer Science at Tulane University.

Dr. Peng's research interests include machine learning, data mining and image databases. He has served as the Guest Co-Editor of the Special Issue on Learning in Computer Vision and Pattern Recognition, IEEE Transactions on Systems, Man, and Cybernetics.  He has been PIs or Co-PIs of various projects from NSF, ARO, ONR, and NASA.  He has authored or co-authored over 100 technical publications in the area of his interest.