Structured Stochastic Variational Inference

Abstract

Stochastic variational inference makes it possible to approximate posterior distributions induced by large datasets quickly using stochastic optimization. The algorithm relies on the use of fully factorized variational distributions. However, this "mean-field" independence approximation limits the fidelity of the posterior approximation, and introduces local optima. We show how to relax the mean-field approximation to allow arbitrary dependencies between global parameters and local hidden variables, producing better parameter estimates by reducing bias, sensitivity to local optima, and sensitivity to hyperparameters.

References in corpus (3)

• On Smoothing and Inference for Topic Models
• Variational Bayesian Inference with Stochastic Search
• Sparse Stochastic Inference for Latent Dirichlet allocation

Cited by in corpus (6)

• Deep Probabilistic Programming
• Adversarial Message Passing For Graphical Models
• A trust-region method for stochastic variational inference with applications to streaming data
• Perturb-and-MPM: Quantifying Segmentation Uncertainty in Dense Multi-Label CRFs
• An Empirical Study of Stochastic Variational Algorithms for the Beta Bernoulli Process
• Random Function Priors for Correlation Modeling