JSM 2015 Preliminary Program

Online Program Home
My Program

Abstract Details

Activity Number: 434
Type: Contributed
Date/Time: Tuesday, August 11, 2015 : 2:00 PM to 3:50 PM
Sponsor: Section on Statistical Learning and Data Mining
Abstract #317513 View Presentation
Title: Submodularity in Statistics
Author(s): Kory Johnson* and Robert Stine and Dean Foster
Companies: The Wharton School and The Wharton School and Yahoo Lab
Keywords: Lasso ; Variable Selection ; Greedy Algorithms ; Submodular ; Restricted Eigenvalue ; Stepwise Regression
Abstract:

We demonstrate the usefulness of submodularity in statistics. Greedy algorithms such as forward stepwise regression and the lasso perform well in situations that can be characterized by submodularity. In particular, submodularity of the coefficient of determination, R$^2$, provides a natural way to analyze the effects of collinearity on model selection. In model selection, we encounter the search problem of identifying a subset of $k$ covariates with predictive loss close to that of the best model of $k$ covariates. Submodularity arises naturally in this setting due to its deep roots within combinatorial optimization. It provides structural results for discrete convexity as well as guarantees for the success of greedy algorithms. In statistics, submodularity isolates cases in which collinearity makes the choice of model features difficult from those in which this task is routine. Submodularity of R$^2$ is closely related to other statistical assumptions used in variable screening and proving the performance of the Lasso. This has important implications for the use of model selection procedures: the situations in which forward stepwise and Lasso are successful are closely related.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2015 program





For program information, contact the JSM Registration Department or phone (888) 231-3473.

For Professional Development information, contact the Education Department.

The views expressed here are those of the individual authors and not necessarily those of the JSM sponsors, their officers, or their staff.

2015 JSM Online Program Home