References & Citations
Statistics > Computation
Title: Hamiltonian Monte Carlo Swindles
(Submitted on 14 Jan 2020 (this version), latest version 2 Mar 2020 (v2))
Abstract: Hamiltonian Monte Carlo (HMC) is a powerful Markov chain Monte Carlo (MCMC) algorithm for estimating expectations with respect to continuous un-normalized probability distributions. MCMC estimators typically have higher variance than classical Monte Carlo with i.i.d. samples due to autocorrelations; most MCMC research tries to reduce these autocorrelations. In this work, we explore a complementary approach to variance reduction based on two classical Monte Carlo "swindles": first, running an auxiliary coupled chain targeting a tractable approximation to the target distribution, and using the auxiliary samples as control variates; and second, generating anti-correlated ("antithetic") samples by running two chains with flipped randomness. Both ideas have been explored previously in the context of Gibbs samplers and random-walk Metropolis algorithms, but we argue that they are ripe for adaptation to HMC in light of recent coupling results from the HMC theory literature. For many posterior distributions, we find that these swindles generate effective sample sizes orders of magnitude larger than plain HMC, as well as being more efficient than analogous swindles for Metropolis-adjusted Langevin algorithm and random-walk Metropolis.
Submission history
From: Pavel Sountsov [view email][v1] Tue, 14 Jan 2020 20:08:20 GMT (293kb,D)
[v2] Mon, 2 Mar 2020 20:34:53 GMT (356kb,D)
Link back to: arXiv, form interface, contact.