Accelerating Approximate Thompson Sampling with Underdamped Langevin Monte Carlo

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Zheng, Haoyang, Deng, Wei, Moya, Christian, Lin, Guang
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866929392754622464
author Zheng, Haoyang
Deng, Wei
Moya, Christian
Lin, Guang
author_facet Zheng, Haoyang
Deng, Wei
Moya, Christian
Lin, Guang
contents Approximate Thompson sampling with Langevin Monte Carlo broadens its reach from Gaussian posterior sampling to encompass more general smooth posteriors. However, it still encounters scalability issues in high-dimensional problems when demanding high accuracy. To address this, we propose an approximate Thompson sampling strategy, utilizing underdamped Langevin Monte Carlo, where the latter is the go-to workhorse for simulations of high-dimensional posteriors. Based on the standard smoothness and log-concavity conditions, we study the accelerated posterior concentration and sampling using a specific potential function. This design improves the sample complexity for realizing logarithmic regrets from $\mathcal{\tilde O}(d)$ to $\mathcal{\tilde O}(\sqrt{d})$. The scalability and robustness of our algorithm are also empirically validated through synthetic experiments in high-dimensional bandit problems.
format Preprint
id arxiv_https___arxiv_org_abs_2401_11665
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Accelerating Approximate Thompson Sampling with Underdamped Langevin Monte Carlo
Zheng, Haoyang
Deng, Wei
Moya, Christian
Lin, Guang
Machine Learning
Artificial Intelligence
Approximate Thompson sampling with Langevin Monte Carlo broadens its reach from Gaussian posterior sampling to encompass more general smooth posteriors. However, it still encounters scalability issues in high-dimensional problems when demanding high accuracy. To address this, we propose an approximate Thompson sampling strategy, utilizing underdamped Langevin Monte Carlo, where the latter is the go-to workhorse for simulations of high-dimensional posteriors. Based on the standard smoothness and log-concavity conditions, we study the accelerated posterior concentration and sampling using a specific potential function. This design improves the sample complexity for realizing logarithmic regrets from $\mathcal{\tilde O}(d)$ to $\mathcal{\tilde O}(\sqrt{d})$. The scalability and robustness of our algorithm are also empirically validated through synthetic experiments in high-dimensional bandit problems.
title Accelerating Approximate Thompson Sampling with Underdamped Langevin Monte Carlo
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2401.11665