|
Activity Number:
|
321
- Machine Learning and Variable Selection
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 11, 2021 : 3:30 PM to 5:20 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
|
Abstract #318662
|
|
|
Title:
|
A Minimum Penalty for Multivariate Regression
|
|
Author(s):
|
Brad Price* and Ben Sherwood
|
|
Companies:
|
John Chambers College of Business and Economics, West Virginia University and School of Business, University of Kansas
|
|
Keywords:
|
Multivariate Regression;
Multi-Task Learning;
Fusion Penalties;
Non-Convex Optimization;
Generalized Linear Models
|
|
Abstract:
|
In this work we present a novel framework for multivariate regression based on the use of the minimum function as a penalty which simultaneously learns the relationships between responses while estimating regression coefficients. The approach uses a novel L2 penalty which The algorithms we propose for both least squares and binomial logistic regression use an iterative approach that is a generalization from other works that try to find groups of coefficients in regression. We present theoretical insights on high dimensional consistency, model selection consistency, and post-selection inference. To show the impact of the proposed methodology we present examples in the fields of genetics and substance abuse/misuse.
|
Authors who are presenting talks have a * after their name.
