Activity Number:
|
541
- Recent Progresses in Bayesian Inference in Large Parameter Spaces: Jayanta K. Ghosh Memorial Session
|
Type:
|
Invited
|
Date/Time:
|
Wednesday, July 31, 2019 : 2:00 PM to 3:50 PM
|
Sponsor:
|
Memorial
|
Abstract #300432
|
Presentation
|
Title:
|
Sorted L-One Penalized Estimation
|
Author(s):
|
Malgorzata Bogdan*
|
Companies:
|
University of Wroclaw
|
Keywords:
|
model selection;
multiple testing;
convex optimization;
False Discovery Rate;
Bayesian statistics
|
Abstract:
|
Sorted L-One Penalized Estimation (SLOPE) is a relatively new convex optimization method for identifying predictors in large data bases. Sorted L-One Penalty reduces the dimension by shrinking regression coefficients to zero as well as by making them equal to each other (i.e. shrinking towards the group mean). It provably allows for FDR control under orthogonal designs and yields asymptotically minimax estimators of regression coefficients in sparse high-dimensional regression. We will briefly introduce the idea of the method and present new theoretical results concerning asymptotic FDR control and grouping properties of SLOPE. We will also present a novel adaptive Bayesian version of SLOPE, which allows for reduction of the bias of estimated regression coefficients and control of False Discovery Rate under a wider range of scenarios as compared to original SLOPE. Theoretical results will be illustrated with computer simulations and real data analysis in the context of the analysis of genetic and financial data.
|
Authors who are presenting talks have a * after their name.
Back to the full JSM 2019 program
|