Iterative Amortized Inference

TL;DR

This paper introduces iterative inference models that improve upon standard amortized inference in VAEs by performing repeated gradient-based updates, effectively closing the amortization gap and enhancing inference quality.

Contribution

It proposes a novel iterative inference approach that generalizes standard models and learns to perform inference optimization through gradient encoding.

Findings

Iterative inference models outperform standard models on image and text benchmarks.

The approach effectively closes the amortization gap in variational inference.

Provides insights into top-down inference techniques.

Abstract

Inference models are a key component in scaling variational inference to deep latent variable models, most notably as encoder networks in variational auto-encoders (VAEs). By replacing conventional optimization-based inference with a learned model, inference is amortized over data examples and therefore more computationally efficient. However, standard inference models are restricted to direct mappings from data to approximate posterior estimates. The failure of these models to reach fully optimized approximate posterior estimates results in an amortization gap. We aim toward closing this gap by proposing iterative inference models, which learn to perform inference optimization through repeatedly encoding gradients. Our approach generalizes standard inference models in VAEs and provides insight into several empirical findings, including top-down inference techniques. We demonstrate the inference optimization capabilities of iterative inference models and show that they outperform standard inference models on several benchmark data sets of images and text.