Auto-Encoding Sequential Monte Carlo

TL;DR

This paper introduces an improved auto-encoding sequential Monte Carlo method that combines SMC efficiency with neural network flexibility for scalable model and proposal learning in deep generative models.

Contribution

It presents new theoretical insights and a novel training procedure enhancing model and proposal learning in the AESMC framework.

Findings

Improved training procedure enhances learning efficiency.

Scalable method for deep generative models.

Demonstrated faster and easier implementation.

Abstract

We build on auto-encoding sequential Monte Carlo (AESMC): a method for model and proposal learning based on maximizing the lower bound to the log marginal likelihood in a broad family of structured probabilistic models. Our approach relies on the efficiency of sequential Monte Carlo (SMC) for performing inference in structured probabilistic models and the flexibility of deep neural networks to model complex conditional probability distributions. We develop additional theoretical insights and introduce a new training procedure which improves both model and proposal learning. We demonstrate that our approach provides a fast, easy-to-implement and scalable means for simultaneous model learning and proposal adaptation in deep generative models.