Algorithms for mean-field variational inference via polyhedral optimization in the Wasserstein space

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jiang, Yiheng, Chewi, Sinho, Pooladian, Aram-Alexandre
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912402716491776
author Jiang, Yiheng
Chewi, Sinho
Pooladian, Aram-Alexandre
author_facet Jiang, Yiheng
Chewi, Sinho
Pooladian, Aram-Alexandre
contents We develop a theory of finite-dimensional polyhedral subsets over the Wasserstein space and optimization of functionals over them via first-order methods. Our main application is to the problem of mean-field variational inference, which seeks to approximate a distribution $π$ over $\mathbb{R}^d$ by a product measure $π^\star$. When $π$ is strongly log-concave and log-smooth, we provide (1) approximation rates certifying that $π^\star$ is close to the minimizer $π^\star_\diamond$ of the KL divergence over a \emph{polyhedral} set $\mathcal{P}_\diamond$, and (2) an algorithm for minimizing $\text{KL}(\cdot\|π)$ over $\mathcal{P}_\diamond$ based on accelerated gradient descent over $\R^d$. As a byproduct of our analysis, we obtain the first end-to-end analysis for gradient-based algorithms for MFVI.
format Preprint
id arxiv_https___arxiv_org_abs_2312_02849
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Algorithms for mean-field variational inference via polyhedral optimization in the Wasserstein space
Jiang, Yiheng
Chewi, Sinho
Pooladian, Aram-Alexandre
Statistics Theory
Machine Learning
Optimization and Control
We develop a theory of finite-dimensional polyhedral subsets over the Wasserstein space and optimization of functionals over them via first-order methods. Our main application is to the problem of mean-field variational inference, which seeks to approximate a distribution $π$ over $\mathbb{R}^d$ by a product measure $π^\star$. When $π$ is strongly log-concave and log-smooth, we provide (1) approximation rates certifying that $π^\star$ is close to the minimizer $π^\star_\diamond$ of the KL divergence over a \emph{polyhedral} set $\mathcal{P}_\diamond$, and (2) an algorithm for minimizing $\text{KL}(\cdot\|π)$ over $\mathcal{P}_\diamond$ based on accelerated gradient descent over $\R^d$. As a byproduct of our analysis, we obtain the first end-to-end analysis for gradient-based algorithms for MFVI.
title Algorithms for mean-field variational inference via polyhedral optimization in the Wasserstein space
topic Statistics Theory
Machine Learning
Optimization and Control
url https://arxiv.org/abs/2312.02849