Cameron Bell, Krzystof Łatuszyński, Gareth O. Roberts
{"title":"自适应立体 MCMC","authors":"Cameron Bell, Krzystof Łatuszyński, Gareth O. Roberts","doi":"arxiv-2408.11780","DOIUrl":null,"url":null,"abstract":"In order to tackle the problem of sampling from heavy tailed, high\ndimensional distributions via Markov Chain Monte Carlo (MCMC) methods, Yang,\nLatuszy\\'nski, and Roberts (2022) (arXiv:2205.12112) introduces the\nstereographic projection as a tool to compactify $\\mathbb{R}^d$ and transform\nthe problem into sampling from a density on the unit sphere $\\mathbb{S}^d$.\nHowever, the improvement in algorithmic efficiency, as well as the\ncomputational cost of the implementation, are still significantly impacted by\nthe parameters used in this transformation. To address this, we introduce adaptive versions of the Stereographic Random\nWalk (SRW), the Stereographic Slice Sampler (SSS), and the Stereographic Bouncy\nParticle Sampler (SBPS), which automatically update the parameters of the\nalgorithms as the run progresses. The adaptive setup allows us to better\nexploit the power of the stereographic projection, even when the target\ndistribution is neither centered nor homogeneous. We present a simulation study\nshowing each algorithm's robustness to starting far from the mean in heavy\ntailed, high dimensional settings, as opposed to Hamiltonian Monte Carlo (HMC).\nWe establish a novel framework for proving convergence of adaptive MCMC\nalgorithms over collections of simultaneously uniformly ergodic Markov\noperators, including continuous time processes. This framework allows us to\nprove LLNs and a CLT for our adaptive Stereographic algorithms.","PeriodicalId":501215,"journal":{"name":"arXiv - STAT - Computation","volume":"14 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-08-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Adaptive Stereographic MCMC\",\"authors\":\"Cameron Bell, Krzystof Łatuszyński, Gareth O. Roberts\",\"doi\":\"arxiv-2408.11780\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"In order to tackle the problem of sampling from heavy tailed, high\\ndimensional distributions via Markov Chain Monte Carlo (MCMC) methods, Yang,\\nLatuszy\\\\'nski, and Roberts (2022) (arXiv:2205.12112) introduces the\\nstereographic projection as a tool to compactify $\\\\mathbb{R}^d$ and transform\\nthe problem into sampling from a density on the unit sphere $\\\\mathbb{S}^d$.\\nHowever, the improvement in algorithmic efficiency, as well as the\\ncomputational cost of the implementation, are still significantly impacted by\\nthe parameters used in this transformation. To address this, we introduce adaptive versions of the Stereographic Random\\nWalk (SRW), the Stereographic Slice Sampler (SSS), and the Stereographic Bouncy\\nParticle Sampler (SBPS), which automatically update the parameters of the\\nalgorithms as the run progresses. The adaptive setup allows us to better\\nexploit the power of the stereographic projection, even when the target\\ndistribution is neither centered nor homogeneous. We present a simulation study\\nshowing each algorithm's robustness to starting far from the mean in heavy\\ntailed, high dimensional settings, as opposed to Hamiltonian Monte Carlo (HMC).\\nWe establish a novel framework for proving convergence of adaptive MCMC\\nalgorithms over collections of simultaneously uniformly ergodic Markov\\noperators, including continuous time processes. This framework allows us to\\nprove LLNs and a CLT for our adaptive Stereographic algorithms.\",\"PeriodicalId\":501215,\"journal\":{\"name\":\"arXiv - STAT - Computation\",\"volume\":\"14 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-08-21\",\"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-2408.11780\",\"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-2408.11780","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
In order to tackle the problem of sampling from heavy tailed, high
dimensional distributions via Markov Chain Monte Carlo (MCMC) methods, Yang,
Latuszy\'nski, and Roberts (2022) (arXiv:2205.12112) introduces the
stereographic projection as a tool to compactify $\mathbb{R}^d$ and transform
the problem into sampling from a density on the unit sphere $\mathbb{S}^d$.
However, the improvement in algorithmic efficiency, as well as the
computational cost of the implementation, are still significantly impacted by
the parameters used in this transformation. To address this, we introduce adaptive versions of the Stereographic Random
Walk (SRW), the Stereographic Slice Sampler (SSS), and the Stereographic Bouncy
Particle Sampler (SBPS), which automatically update the parameters of the
algorithms as the run progresses. The adaptive setup allows us to better
exploit the power of the stereographic projection, even when the target
distribution is neither centered nor homogeneous. We present a simulation study
showing each algorithm's robustness to starting far from the mean in heavy
tailed, high dimensional settings, as opposed to Hamiltonian Monte Carlo (HMC).
We establish a novel framework for proving convergence of adaptive MCMC
algorithms over collections of simultaneously uniformly ergodic Markov
operators, including continuous time processes. This framework allows us to
prove LLNs and a CLT for our adaptive Stereographic algorithms.