From Margin-Based Classification to psi-learning

• Activity Number: 159
• Type: Invited
• Date/Time: Monday, August 4, 2003 : 2:00 PM to 3:50 PM
• Sponsor: IMS
• Author(s): Xiaotong Shen*+
• Companies: Ohio State University
• Keywords: classification ; support vectors ; d.c. programming

Abstract

The concept of large margins plays an important role in analyzing learning methodologies such as Boosting, Neural Networks, and Support Vector Machines. In this talk, I will present a new learning methodology called psi-learning as well as the associated computational tools. While retaining the interpretation of large margins, $\psi$-learning delivers high performance in generalization, especially in nonseparable cases, as it is derived from a direct consideration of generalization errors. In order to realize its potential, we tackle a difficult nonconvex minimization problem, utilizing the global optimization techniques. In particular, we propose two computational strategies based on d.c. (differenced convex) programming, yielding a sequential quadratic program for the minimization problem. Numerical experiments are performed using simulated and benchmark examples, which suggest that the computational strategies can realize the theoretical advantages of psi-learning.