Stochastic Stein Discrepancies
arXiv:2007.02857
Abstract
Stein discrepancies (SDs) monitor convergence and non-convergence in approximate inference when exact integration and sampling are intractable. However, the computation of a Stein discrepancy can be prohibitive if the Stein operator - often a sum over likelihood terms or potentials - is expensive to evaluate. To address this deficiency, we show that stochastic Stein discrepancies (SSDs) based on subsampled approximations of the Stein operator inherit the convergence control properties of standard SDs with probability 1. Along the way, we establish the convergence of Stein variational gradient descent (SVGD) on unbounded domains, resolving an open question of Liu (2017). In our experiments with biased Markov chain Monte Carlo (MCMC) hyperparameter tuning, approximate MCMC sampler selection, and stochastic SVGD, SSDs deliver comparable inferences to standard SDs with orders of magnitude fewer likelihood evaluations.
References in corpus (13)
- SAGA: A Fast Incremental Gradient Method With Support for Non-Strongly Convex Composite Objectives
- Stochastic Gradient Hamiltonian Monte Carlo
- Stein Variational Gradient Descent: A General Purpose Bayesian Inference Algorithm
- Stein Variational Gradient Descent as Gradient Flow
- Julia: A Fresh Approach to Numerical Computing
- Measuring Sample Quality with Kernels
- Learning to Draw Samples: With Application to Amortized MLE for Generative Adversarial Learning
- Zero Variance Markov Chain Monte Carlo for Bayesian Estimators
- Message Passing Stein Variational Gradient Descent
- A stochastic version of Stein Variational Gradient Descent for efficient sampling
- Random Feature Stein Discrepancies
- Stein Point Markov Chain Monte Carlo
- Control functionals for Monte Carlo integration