Congye Wang, Wilson Chen, Heishiro Kanagawa, Chris. J. Oates
{"title":"自适应 MCMC 的强化学习","authors":"Congye Wang, Wilson Chen, Heishiro Kanagawa, Chris. J. Oates","doi":"arxiv-2405.13574","DOIUrl":null,"url":null,"abstract":"An informal observation, made by several authors, is that the adaptive design\nof a Markov transition kernel has the flavour of a reinforcement learning task.\nYet, to-date it has remained unclear how to actually exploit modern\nreinforcement learning technologies for adaptive MCMC. The aim of this paper is\nto set out a general framework, called Reinforcement Learning\nMetropolis--Hastings, that is theoretically supported and empirically\nvalidated. Our principal focus is on learning fast-mixing Metropolis--Hastings\ntransition kernels, which we cast as deterministic policies and optimise via a\npolicy gradient. Control of the learning rate provably ensures conditions for\nergodicity are satisfied. The methodology is used to construct a gradient-free\nsampler that out-performs a popular gradient-free adaptive Metropolis--Hastings\nalgorithm on $\\approx 90 \\%$ of tasks in the PosteriorDB benchmark.","PeriodicalId":501215,"journal":{"name":"arXiv - STAT - Computation","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-05-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Reinforcement Learning for Adaptive MCMC\",\"authors\":\"Congye Wang, Wilson Chen, Heishiro Kanagawa, Chris. J. Oates\",\"doi\":\"arxiv-2405.13574\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"An informal observation, made by several authors, is that the adaptive design\\nof a Markov transition kernel has the flavour of a reinforcement learning task.\\nYet, to-date it has remained unclear how to actually exploit modern\\nreinforcement learning technologies for adaptive MCMC. The aim of this paper is\\nto set out a general framework, called Reinforcement Learning\\nMetropolis--Hastings, that is theoretically supported and empirically\\nvalidated. Our principal focus is on learning fast-mixing Metropolis--Hastings\\ntransition kernels, which we cast as deterministic policies and optimise via a\\npolicy gradient. Control of the learning rate provably ensures conditions for\\nergodicity are satisfied. The methodology is used to construct a gradient-free\\nsampler that out-performs a popular gradient-free adaptive Metropolis--Hastings\\nalgorithm on $\\\\approx 90 \\\\%$ of tasks in the PosteriorDB benchmark.\",\"PeriodicalId\":501215,\"journal\":{\"name\":\"arXiv - STAT - Computation\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-05-22\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - STAT - Computation\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2405.13574\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - STAT - Computation","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2405.13574","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
An informal observation, made by several authors, is that the adaptive design
of a Markov transition kernel has the flavour of a reinforcement learning task.
Yet, to-date it has remained unclear how to actually exploit modern
reinforcement learning technologies for adaptive MCMC. The aim of this paper is
to set out a general framework, called Reinforcement Learning
Metropolis--Hastings, that is theoretically supported and empirically
validated. Our principal focus is on learning fast-mixing Metropolis--Hastings
transition kernels, which we cast as deterministic policies and optimise via a
policy gradient. Control of the learning rate provably ensures conditions for
ergodicity are satisfied. The methodology is used to construct a gradient-free
sampler that out-performs a popular gradient-free adaptive Metropolis--Hastings
algorithm on $\approx 90 \%$ of tasks in the PosteriorDB benchmark.