|
Activity Number:
|
357
- Contemporary Multivariate Methods
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2020 : 10:00 AM to 2:00 PM
|
|
Sponsor:
|
Section on Statistical Learning and Data Science
|
|
Abstract #312743
|
|
|
Title:
|
Stepsize Selection in Langevin Monte Carlo via Coupling
|
|
Author(s):
|
Matteo Sordello* and Weijie Su and James Johndrow
|
|
Companies:
|
University of Pennsylvania and University of Pennsylvania and University of Pennsylvania
|
|
Keywords:
|
Langevin;
Monte Carlo;
sampling;
coupling
|
|
Abstract:
|
In this paper we propose a new sampling procedure with a dynamic stepsize selection rule. We use Langevin Monte Carlo to approximately sample from a target distribution, and adaptively decrease the stepsize when stationarity is detected. The detection is performed by running two coupled chains with different starting points, and waiting for their coupling time. We perform experiments with unimodal and multimodal target distributions, and compare the performance of this procedure with other state-of-the-art algorithms.
|
Authors who are presenting talks have a * after their name.
