Rate-optimal refinement strategies for local approximation MCMC
نویسندگان
چکیده
Many Bayesian inference problems involve target distributions whose density functions are computationally expensive to evaluate. Replacing the with a local approximation based on small number of carefully chosen evaluations can significantly reduce computational expense Markov chain Monte Carlo (MCMC) sampling. Moreover, continual refinement guarantee asymptotically exact We devise new strategy for balancing decay rate bias due that MCMC variance. prove error resulting (LA-MCMC) algorithm decays at roughly expected $$1/\sqrt{T}$$ rate, and we demonstrate this numerically. also introduce an algorithmic parameter guarantees convergence given very weak tail bounds, strengthening previous results. Finally, apply LA-MCMC intensive inverse problem arising in groundwater hydrology.
منابع مشابه
Partition of Unity Refinement for local approximation
In this article, we propose a Partition ofUnity Refinement (PUR)method to improve the local approximations of elliptic boundary value problems in regions of interest. The PUR method only needs to refine the local meshes and hanging nodes generate no difficulty. The mesh qualities such as uniformity or quasi-uniformity are kept. The advantages of the PUR include its effectiveness and relatively ...
متن کاملSampling Strategies for MCMC
There are many good methods for sampling Markov chains via streams of independent U [0, 1] random variables. Recently some non-random and some random but dependent driving sequences have been shown to result in consistent Markov chain sampling, sometimes with considerably improved accuracy. The key to consistent sampling is for the driving sequence to be completely uniformly distributed (CUD) o...
متن کاملBasis refinement strategies for linear value function approximation in MDPs
We provide a theoretical framework for analyzing basis function construction for linear value function approximation in Markov Decision Processes (MDPs). We show that important existing methods, such as Krylov bases and Bellman-errorbased methods are a special case of the general framework we develop. We provide a general algorithmic framework for computing basis function refinements which “res...
متن کاملLocal Bandit Approximation for Optimal Learning Problems
In general, procedures for determining Bayes-optimal adaptive controls for Markov decision processes (MDP's) require a prohibitive amount of computation-the optimal learning problem is intractable. This paper proposes an approximate approach in which bandit processes are used to model, in a certain "local" sense, a given MDP. Bandit processes constitute an important subclass of MDP's, and have ...
متن کاملControlled MCMC for Optimal Sampling
In this paper we develop an original and general framework for automatically optimizing the statistical properties of Markov chain Monte Carlo (MCMC) samples, which are typically used to evaluate complex integrals. The Metropolis-Hastings algorithm is the basic building block of classical MCMC methods and requires the choice of a proposal distribution, which usually belongs to a parametric fami...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Statistics and Computing
سال: 2022
ISSN: ['0960-3174', '1573-1375']
DOI: https://doi.org/10.1007/s11222-022-10123-0