High-Dimensional Learning for Ordinal and Multiclass Data

• Activity Number: 299
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Xingye Qiao*+
• Companies: Binghamton University
• Keywords: Fisher consistency ; High-dimensional, low-sample size data ; Multiclass classification ; Non-crossing constraints ; Ordinal classification ; Statistical learning

Abstract

In this talk, I will discuss two projects. In the first project, the goal is to classify ordinal data, which have class labels from a set of categories with inherent orders. We propose a non-crossing ordinal classifier where multiple binary ones are trained simultaneously with a constraint that rules out crossing phenomena among their classification boundaries. In the second project, the aim is to conduct variable selection for multiclass classification. With the state-of-the-art Lasso type of penalization, the resulting variable selection method actually selects coefficients instead of variables. We use three different group penalties to conduct true variable selection for multiclass classification.