Current browse context:
cs.LG
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: Improving Sampling Accuracy of Stochastic Gradient MCMC Methods via Non-uniform Subsampling of Gradients
(Submitted on 20 Feb 2020 (v1), revised 16 Feb 2021 (this version, v2), latest version 22 Sep 2021 (v3))
Abstract: Stochastic Gradient (SG-)MCMC methods for sampling statistical distributions approximate gradients by stochastic ones, commonly via uniformly subsampled data points. We propose a non-uniform subsampling scheme to improve the sampling accuracy. The proposed exponentially weighted stochastic gradient (EWSG) is designed so that a non-uniform-SG-MCMC method mimics the statistical behavior of a batch-gradient-MCMC method, and hence the inaccuracy due to SG approximation is reduced. EWSG differs from Variance Reduction (VR) techniques as it focuses on the entire distribution instead of just the variance; nevertheless, its reduced local variance is also proved. EWSG can also be viewed as an extension of the importance sampling idea, successful for SG-based optimizations, to sampling tasks. In our practical implementation of EWSG, the non-uniform subsampling is performed efficiently via a Metropolis-Hasting chain on the data index, which is coupled to the MCMC algorithm. Numerical experiments are provided, not only to demonstrate EWSG's effectiveness, but also to guide hyperparameter choices, and validate our \emph{non-asymptotic global error bound} despite of approximations in the implementation. Notably, while statistical accuracy is improved, convergence speed can be comparable to the uniform version, which renders EWSG a practical alternative to VR (but EWSG and VR can be combined too).
Submission history
From: Molei Tao [view email][v1] Thu, 20 Feb 2020 18:56:18 GMT (555kb,D)
[v2] Tue, 16 Feb 2021 17:12:34 GMT (1662kb,D)
[v3] Wed, 22 Sep 2021 22:49:34 GMT (6183kb,D)
Link back to: arXiv, form interface, contact.