|
Activity Number:
|
355
- Advanced Bayesian Topics (Part 4)
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 12, 2021 : 10:00 AM to 11:50 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
|
Abstract #318768
|
|
|
Title:
|
Nonparametric Empirical Bayes Estimation with Entropic Optimal Transport
|
|
Author(s):
|
Jordan Grey Bryan*
|
|
Companies:
|
Department of Statistical Science, Duke University
|
|
Keywords:
|
Empirical Bayes;
Optimal transport
|
|
Abstract:
|
We revisit the classical empirical Bayes estimation procedure for a sample corrupted by noise. Specifically, we focus on an estimator obtained by making minimal assumptions on the prior distribution from which the uncorrupted random variables are drawn. We show that maximizing the marginal data likelihood under this unparameterized prior coincides with solving an entropic optimal transport problem. Using tools developed for other transport-based estimators, we analyze the risk properties of the Bayes-optimal estimator derived from this prior distribution and compare them to those of existing empirical Bayes estimators.
|
Authors who are presenting talks have a * after their name.
