Unbiased Estimation using a Class of Diffusion Processes

DOI10.1016/J.JCP.2022.111643MaRDI QIDQ6392941arXivFDO

Authors Hamza M. Ruzayqat, Alexandros Beskos, Dan Crisan, Ajay Jasra, Nikolas Kantas

Publication date 6 March 2022

Published in arXiv

Full work available at URL https://doi.org/10.1016/J.JCP.2022.111643

Mathematics Subject Classification ID

Stochastic analysis (60Hxx) Probabilistic methods, stochastic differential equations (65Cxx) Numerical methods for ordinary differential equations (65Lxx)

arXiv classification

cs.NAmath.NAmath.PRstat.COstat.ME

Abstract: We study the problem of unbiased estimation of expectations with respect to (w.r.t.) $pi$ a given, general probability measure on $(mathbb{R}^d,mathcalB(mathbb{R}^d))$ that is absolutely continuous with respect to a standard Gaussian measure. We focus on simulation associated to a particular class of diffusion processes, sometimes termed the Schr"odinger-F"ollmer Sampler, which is a simulation technique that approximates the law of a particular diffusion bridge process ${X_t}_{tin[0,1]}$ on $mathbb{R}^d$, $dinmathbb{N}_0$. This latter process is constructed such that, starting at $X_0=0$, one has $X_{1}simpi$. Typically, the drift of the diffusion is intractable and, even if it were not, exact sampling of the associated diffusion is not possible. As a result, cite{sf_orig,jiao} consider a stochastic Euler-Maruyama scheme that allows the development of biased estimators for expectations w.r.t.~$pi$. We show that for this methodology to achieve a mean square error of $mathcalO(epsilo{n}^2)$, for arbitrary $epsilon>0$, the associated cost is $mathcalO(epsilo{n}^{-5})$. We then introduce an alternative approach that provides unbiased estimates of expectations w.r.t.~$pi$, that is, it does not suffer from the time discretization bias or the bias related with the approximation of the drift function. We prove that to achieve a mean square error of $mathcalO(epsilo{n}^2)$, the associated cost is, with high probability, $mathcalO(epsilo{n}^{-2}|log(epsilon){|}^{2+delta})$, for any $delta>0$. We implement our method on several examples including Bayesian inverse problems.

This page was built for publication: Unbiased Estimation using a Class of Diffusion Processes

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6392941)