Natural Evolution Strategies as a Black Box Estimator for Stochastic Variational Inference

TL;DR

This paper introduces a natural evolution strategies-based estimator for stochastic variational inference, enabling Bayesian inference on complex models without the limitations of the reparameterization trick.

Contribution

It proposes a novel estimator that broadens the scope of models for stochastic variational inference by removing distributional assumptions.

Findings

Allows inference on models incompatible with reparameterization

Provides unbiased gradient estimates without distribution constraints

Enables more flexible Bayesian modeling

Abstract

Stochastic variational inference and its derivatives in the form of variational autoencoders enjoy the ability to perform Bayesian inference on large datasets in an efficient manner. However, performing inference with a VAE requires a certain design choice (i.e. reparameterization trick) to allow unbiased and low variance gradient estimation, restricting the types of models that can be created. To overcome this challenge, an alternative estimator based on natural evolution strategies is proposed. This estimator does not make assumptions about the kind of distributions used, allowing for the creation of models that would otherwise not have been possible under the VAE framework.