Estimating Normalizing Constants for Log-Concave Distributions: Algorithms and Lower Bounds

DOI10.1145/3357713.3384289arXiv1911.03043MaRDI QIDQ6328752FDOQ6328752

Authors: Rong Ge, Holden Lee, Jianfeng Lu 

Publication date: 7 November 2019

Abstract: Estimating the normalizing constant of an unnormalized probability distribution has important applications in computer science, statistical physics, machine learning, and statistics. In this work, we consider the problem of estimating the normalizing constant $Z=in{t}_{mathbb{R}^d}{e}^{-f(x)},mathrmdx$ to within a multiplication factor of $1pmvarepsilon$ for a $mu$-strongly convex and $L$-smooth function $f$, given query access to $f(x)$ and $ablaf(x)$. We give both algorithms and lowerbounds for this problem. Using an annealing algorithm combined with a multilevel Monte Carlo method based on underdamped Langevin dynamics, we show that $widetildemathcalOBigl(frac{d}^{4/3}kappa+d^{7/6}kapp{a}^{7/6}varepsilo{n}^{2}Bigr)$ queries to $ablaf$ are sufficient, where $kappa=L/mu$ is the condition number. Moreover, we provide an information theoretic lowerbound, showing that at least $frac{d}^{1-o(1)}varepsilo{n}^{2-o(1)}$ queries are necessary. This provides a first nontrivial lowerbound for the problem.

Mathematics Subject Classification ID

Theory of computing (68Qxx)

This page was built for publication: Estimating Normalizing Constants for Log-Concave Distributions: Algorithms and Lower Bounds

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6328752)