Pathwise Derivatives for Multivariate Distributions
arXiv:1806.01856
Abstract
We exploit the link between the transport equation and derivatives of expectations to construct efficient pathwise gradient estimators for multivariate distributions. We focus on two main threads. First, we use null solutions of the transport equation to construct adaptive control variates that can be used to construct gradient estimators with reduced variance. Second, we consider the case of multivariate mixture distributions. In particular we show how to compute pathwise derivatives for mixtures of multivariate Normal distributions with arbitrary means and diagonal covariances. We demonstrate in a variety of experiments in the context of variational inference that our gradient estimators can outperform other methods, especially in high dimensions.
To appear at AISTATS 2019; 16 pages
References in corpus (13)
- The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables
- Modeling Temporal Dependencies in High-Dimensional Sequences: Application to Polyphonic Music Generation and Transcription
- REBAR: Low-variance, unbiased gradient estimates for discrete latent variable models
- Implicit Reparameterization Gradients
- Variational Bayesian Inference with Stochastic Search
- Backpropagation through the Void: Optimizing control variates for black-box gradient estimation
- Automated Variational Inference in Probabilistic Programming
- Sticking the Landing: Simple, Lower-Variance Gradient Estimators for Variational Inference
- Variational Boosting: Iteratively Refining Posterior Approximations
- Pathwise Derivatives Beyond the Reparameterization Trick
- Reducing Reparameterization Gradient Variance
- Stochastic Backpropagation through Mixture Density Distributions
- Quadrature Compound: An approximating family of distributions