Events and Seminars : 2014 Seminars

DEVELOPMENTS USING FUSION PENALTIES IN STATISTICAL LEARNING

BRAD PRICE, PH.D.
Clinical Assistant Professor
Department of Management Science
School of Business Administration
University of Miami
TUESDAY, OCTOBER 7, 2014
11:00 a.m.– 12:00 p.m., CRB 692

In this work we propose penalized likelihood estimators for use in statistical classification problems. Our main focus is on the development of a specific type of penalty function which are called fusion penalties. We will discuss developments of fusion penalized likelihood estimators for use in quadratic discriminant analysis and multinomial logistic regression. This first method we discuss is the ridge fusion method which is used to jointly estimate inverse covariance matrices for specific use in quadratic discriminant analysis. The second method we will focus on, group fused multinomial logistic regression, is a new method used to reduce the number of response categories in multinomial logistic regression. Tuning parameter selection for both methods is also addressed.