Bayesian Posterior Sampling via Stochastic Gradient Fisher Scoring

Abstract

In this paper we address the following question: Can we approximately sample from a Bayesian posterior distribution if we are only allowed to touch a small mini-batch of data-items for every sample we generate?. An algorithm based on the Langevin equation with stochastic gradients (SGLD) was previously proposed to solve this, but its mixing rate was slow. By leveraging the Bayesian Central Limit Theorem, we extend the SGLD algorithm so that at high mixing rates it will sample from a normal approximation of the posterior, while for slow mixing rates it will mimic the behavior of SGLD with a pre-conditioner matrix. As a bonus, the proposed algorithm is reminiscent of Fisher scoring (with stochastic gradients) and as such an efficient optimizer during burn-in.

Appears in Proceedings of the 29th International Conference on Machine Learning (ICML 2012)

Cited by in corpus (45)

• Variational Inference with Normalizing Flows
• Stochastic Gradient Hamiltonian Monte Carlo
• Variational Dropout and the Local Reparameterization Trick
• Bayesian Dark Knowledge
• Austerity in MCMC Land: Cutting the Metropolis-Hastings Budget
• Preconditioned Stochastic Gradient Langevin Dynamics for Deep Neural Networks
• Measuring Sample Quality with Stein's Method
• Privacy for Free: Posterior Sampling and Stochastic Gradient Monte Carlo
• Structured and Efficient Variational Deep Learning with Matrix Gaussian Posteriors
• Measuring Sample Quality with Kernels
• Expectation propagation as a way of life: A framework for Bayesian inference on partitioned data
• A Survey of Optimization Methods from a Machine Learning Perspective
• Can You Trust This Prediction? Auditing Pointwise Reliability After Learning
• On nonnegative unbiased estimators
• Global Convergence of Stochastic Gradient Hamiltonian Monte Carlo for Non-Convex Stochastic Optimization: Non-Asymptotic Performance Bounds and Momentum-Based Acceleration
• Scalable MCMC for Mixed Membership Stochastic Blockmodels
• Online Bayesian Passive-Aggressive Learning
• Advances in Variational Inference
• Adaptive Meta-Learning Stochastic Gradient Hamiltonian Monte Carlo Simulation for Bayesian Updating of Structural Dynamic Models
• On Last-Layer Algorithms for Classification: Decoupling Representation from Uncertainty Estimation
• Global Convergence of Langevin Dynamics Based Algorithms for Nonconvex Optimization
• Covariance-Controlled Adaptive Langevin Thermostat for Large-Scale Bayesian Sampling
• Adaptive sampling-based quadrature rules for efficient Bayesian prediction
• Nonasymptotic analysis of Stochastic Gradient Hamiltonian Monte Carlo under local conditions for nonconvex optimization
• Simplest random walk for approximating Robin boundary value problems and ergodic limits of reflected diffusions
• On Stationary-Point Hitting Time and Ergodicity of Stochastic Gradient Langevin Dynamics
• Learning Deep Generative Models with Doubly Stochastic MCMC
• A Contour Stochastic Gradient Langevin Dynamics Algorithm for Simulations of Multi-modal Distributions
• Data Augmentation for Bayesian Deep Learning
• Bayesian Inference Forgetting
• Stochastic Gradient Hamiltonian Monte Carlo with Variance Reduction for Bayesian Inference
• Exploiting the Statistics of Learning and Inference
• Stochastic Variance-Reduced Hamilton Monte Carlo Methods
• Uniform minorization condition and convergence bounds for discretizations of kinetic Langevin dynamics
• Schr{ö}dinger-F{ö}llmer Sampler: Sampling without Ergodicity
• Sparse Variational Inference: Bayesian Coresets from Scratch
• Bayesian Coresets: Revisiting the Nonconvex Optimization Perspective
• A fast asynchronous MCMC sampler for sparse Bayesian inference
• Stochastic Discriminative EM
• Secure and Differentially Private Bayesian Learning on Distributed Data
• Structured Stochastic Gradient MCMC
• A Divergence Bound for Hybrids of MCMC and Variational Inference and an Application to Langevin Dynamics and SGVI
• Revisiting the Effects of Stochasticity for Hamiltonian Samplers
• An Algorithm for Distributed Bayesian Inference in Generalized Linear Models
• Mirrored Langevin Dynamics