Measuring Sample Quality with Stein's Method
arXiv:1506.03039
Abstract
To improve the efficiency of Monte Carlo estimation, practitioners are turning to biased Markov chain Monte Carlo procedures that trade off asymptotic exactness for computational speed. The reasoning is sound: a reduction in variance due to more rapid sampling can outweigh the bias introduced. However, the inexactness creates new challenges for sampler and parameter selection, since standard measures of sample quality like effective sample size do not account for asymptotic bias. To address these challenges, we introduce a new computable quality measure based on Stein's method that quantifies the maximum discrepancy between sample and target expectations over a large class of test functions. We use our tool to compare exact, biased, and deterministic sample sequences and illustrate applications to hyperparameter selection, convergence rate assessment, and quantifying bias-variance tradeoffs in posterior inference.
17 pages, 6 figures
References in corpus (6)
- Computing in Operations Research using Julia
- Fast Construction of Nets in Low Dimensional Metrics, and Their Applications
- Bayesian Posterior Sampling via Stochastic Gradient Fisher Scoring
- On the Equivalence between Herding and Conditional Gradient Algorithms
- Control functionals for Monte Carlo integration
- Stein's method of exchangeable pairs for the Beta distribution and generalizations
Cited by in corpus (50)
- Stein Variational Gradient Descent: A General Purpose Bayesian Inference Algorithm
- The Born Supremacy: Quantum Advantage and Training of an Ising Born Machine
- An introduction to sampling via measure transport
- A Kernel Test of Goodness of Fit
- How to Train Your Energy-Based Models
- Measuring Sample Quality with Kernels
- Monte Carlo Gradient Estimation in Machine Learning
- A-NICE-MC: Adversarial Training for MCMC
- Gradient Estimators for Implicit Models
- A Spectral Approach to Gradient Estimation for Implicit Distributions
- A Linear-Time Kernel Goodness-of-Fit Test
- Stein Variational Gradient Descent Without Gradient
- Stein Variational Adaptive Importance Sampling
- Stein Neural Sampler
- Minimum Stein Discrepancy Estimators
- Characterizations of non-normalized discrete probability distributions and their application in statistics
- Sliced Kernelized Stein Discrepancy
- Stein operators, kernels and discrepancies for multivariate continuous distributions
- A Kernelized Stein Discrepancy for Goodness-of-fit Tests and Model Evaluation
- Estimating Convergence of Markov chains with L-Lag Couplings
- Two Methods For Wild Variational Inference
- Validated Variational Inference via Practical Posterior Error Bounds
- Control functionals for Monte Carlo integration
- Minimum -distance estimators for non-normalized parametric models
- Mutual Information Gradient Estimation for Representation Learning
- Kernel-based ANOVA decomposition and Shapley effects -- Application to global sensitivity analysis
- Optimal quantisation of probability measures using maximum mean discrepancy
- An Easy to Interpret Diagnostic for Approximate Inference: Symmetric Divergence Over Simulations
- Blindness of score-based methods to isolated components and mixing proportions
- On Energy-Based Models with Overparametrized Shallow Neural Networks
- Kernelized Complete Conditional Stein Discrepancy
- AIDE: An algorithm for measuring the accuracy of probabilistic inference algorithms
- Stein Variational Gaussian Processes
- Universal Hypothesis Testing with Kernels: Asymptotically Optimal Tests for Goodness of Fit
- A Stein Goodness-of-fit Test for Directional Distributions
- Understanding MCMC Dynamics as Flows on the Wasserstein Space
- A Unifying and Canonical Description of Measure-Preserving Diffusions
- Sampling with Mirrored Stein Operators
- Measure Transport with Kernel Stein Discrepancy
- Nonparametric Score Estimators
- Active Slices for Sliced Stein Discrepancy
- Asymptotically Optimal One- and Two-Sample Testing with Kernels
- Scalable Approximate Inference and Some Applications
- A Stein Goodness of fit Test for Exponential Random Graph Models
- Adaptive MCMC via Combining Local Samplers
- Interpretable Stein Goodness-of-fit Tests on Riemannian Manifolds
- Separation Results between Fixed-Kernel and Feature-Learning Probability Metrics
- Adversarial Stein Training for Graph Energy Models
- Minimum Discrepancy Methods in Uncertainty Quantification
- Measuring Sample Quality in Algorithms for Intractable Normalizing Function Problems