29
0

A New Framework for Variance-Reduced Hamiltonian Monte Carlo

Abstract

We propose a new framework of variance-reduced Hamiltonian Monte Carlo (HMC) methods for sampling from an LL-smooth and mm-strongly log-concave distribution, based on a unified formulation of biased and unbiased variance reduction methods. We study the convergence properties for HMC with gradient estimators which satisfy the Mean-Squared-Error-Bias (MSEB) property. We show that the unbiased gradient estimators, including SAGA and SVRG, based HMC methods achieve highest gradient efficiency with small batch size under high precision regime, and require O~(N+κ2d12ε1+N23κ43d13ε23)\tilde{O}(N + \kappa^2 d^{\frac{1}{2}} \varepsilon^{-1} + N^{\frac{2}{3}} \kappa^{\frac{4}{3}} d^{\frac{1}{3}} \varepsilon^{-\frac{2}{3}} ) gradient complexity to achieve ϵ\epsilon-accuracy in 2-Wasserstein distance. Moreover, our HMC methods with biased gradient estimators, such as SARAH and SARGE, require O~(N+Nκ2d12ε1)\tilde{O}(N+\sqrt{N} \kappa^2 d^{\frac{1}{2}} \varepsilon^{-1}) gradient complexity, which has the same dependency on condition number κ\kappa and dimension dd as full gradient method, but improves the dependency of sample size NN for a factor of N12N^\frac{1}{2}. Experimental results on both synthetic and real-world benchmark data show that our new framework significantly outperforms the full gradient and stochastic gradient HMC approaches. The earliest version of this paper was submitted to ICML 2020 with three weak accept but was not finally accepted.

View on arXiv
Comments on this paper