Search for: All records

Black-box variational inference algorithms use stochastic sampling to analyze diverse statistical models, like those expressed in probabilistic programming languages, without model-specific derivations. While the popular score-function estimator computes unbiased gradient estimates, its variance is often unacceptably large, especially in models with discrete latent variables. We propose a stochastic natural gradient estimator that is as broadly applicable and unbiased, but improves efficiency by exploiting the curvature of the variational bound, and provably reduces variance by marginalizing discrete latent variables. Our marginalized stochastic natural gradients have intriguing connections to classic coordinate ascent variational inference, but allow parallel updates of variational parameters, and provide superior convergence guarantees relative to naive Monte Carlo approximations. We integrate our method with the probabilistic programming language Pyro and evaluate real-world models of documents, images, networks, and crowd-sourcing. Compared to score-function estimators, we require far fewer Monte Carlo samples and consistently convergence orders of magnitude faster. 

We propose a broadly applicable variational inference algorithm for probabilistic models with binary latent variables, using sampling to approximate expectations required for coordinate ascent updates. Applied to three real-world models for text and image and network data, our approach converges much faster than REINFORCE-style stochastic gradient algorithms, and requires fewer Monte Carlo samples. Compared to hand-crafted variational bounds with model-dependent auxiliary variables, our approach leads to tighter likelihood bounds and greater robustness to local optima. Our method is designed to integrate easily with probabilistic programming languages for effective, scalable, black-box variational inference.