Measuring Sample Quality with Kernels

Abstract

Approximate Markov chain Monte Carlo (MCMC) offers the promise of more rapid sampling at the cost of more biased inference. Since standard MCMC diagnostics fail to detect these biases, researchers have developed computable Stein discrepancy measures that provably determine the convergence of a sample to its target distribution. This approach was recently combined with the theory of reproducing kernels to define a closed-form kernel Stein discrepancy (KSD) computable by summing kernel evaluations across pairs of sample points. We develop a theory of weak convergence for KSDs based on Stein's method, demonstrate that commonly used KSDs fail to detect non-convergence even for Gaussian targets, and show that kernels with slowly decaying tails provably determine convergence for a large class of target distributions. The resulting convergence-determining KSDs are suitable for comparing biased, exact, and deterministic sample sequences and simpler to compute and parallelize than alternative Stein discrepancies. We use our tools to compare biased samplers, select sampler hyperparameters, and improve upon existing KSD approaches to one-sample hypothesis testing and sample quality improvement.

References in corpus (3)

• Bayesian Posterior Sampling via Stochastic Gradient Fisher Scoring
• Learning to Draw Samples: With Application to Amortized MLE for Generative Adversarial Learning
• Two Methods For Wild Variational Inference

Cited by in corpus (36)

• A-NICE-MC: Adversarial Training for MCMC
• Gradient Estimators for Implicit Models
• A Linear-Time Kernel Goodness-of-Fit Test
• Learning the Stein Discrepancy for Training and Evaluating Energy-Based Models without Sampling
• A Non-Asymptotic Analysis for Stein Variational Gradient Descent
• Stochastic Runge-Kutta Accelerates Langevin Monte Carlo and Beyond
• Stein Variational Gradient Descent Without Gradient
• Metrizing Weak Convergence with Maximum Mean Discrepancies
• Practical bounds on the error of Bayesian posterior approximations: A nonasymptotic approach
• Minimum Stein Discrepancy Estimators
• Stein Neural Sampler
• On the Optimality of Kernel-Embedding Based Goodness-of-Fit Tests
• The reproducing Stein kernel approach for post-hoc corrected sampling
• Stein operators, kernels and discrepancies for multivariate continuous distributions
• Geometric Methods for Sampling, Optimisation, Inference and Adaptive Agents
• Stochastic Stein Discrepancies
• Path Integral Sampler: a stochastic control approach for sampling
• Optimal quantisation of probability measures using maximum mean discrepancy
• On Energy-Based Models with Overparametrized Shallow Neural Networks
• Gibbs Sampling the Posterior of Neural Networks
• Blindness of score-based methods to isolated components and mixing proportions
• Kernelized Complete Conditional Stein Discrepancy
• Convergence of Stein Variational Gradient Descent under a Weaker Smoothness Condition
• Universal Hypothesis Testing with Kernels: Asymptotically Optimal Tests for Goodness of Fit
• Posterior Integration on a Riemannian Manifold
• Sampling with Mirrored Stein Operators
• Testing Goodness of Fit of Conditional Density Models with Kernels
• Stein-type covariance identities: Klaassen, Papathanasiou and Olkin-Shepp type bounds for arbitrary target distributions
• Fast Convergence of Langevin Dynamics on Manifold: Geodesics meet Log-Sobolev
• Variational (Gradient) Estimate of the Score Function in Energy-based Latent Variable Models
• Active Slices for Sliced Stein Discrepancy
• Scalable Approximate Inference and Some Applications
• Asymptotically Optimal One- and Two-Sample Testing with Kernels
• Adaptive MCMC via Combining Local Samplers
• Stein Variational Goal Generation for adaptive Exploration in Multi-Goal Reinforcement Learning
• Measuring Sample Quality in Algorithms for Intractable Normalizing Function Problems