A Recurrent Latent Variable Model for Sequential Data

TL;DR

This paper introduces a variational recurrent neural network (VRNN) that incorporates latent variables into the hidden state to better model variability in structured sequential data like speech and handwriting.

Contribution

It presents a novel VRNN model that combines variational autoencoder principles with RNNs, enhancing their ability to capture complex data variability.

Findings

VRNN outperforms related models on speech and handwriting datasets.

Latent variables significantly improve modeling of variability in sequential data.

The approach demonstrates the importance of high-level latent variables in RNNs.

Abstract

In this paper, we explore the inclusion of latent random variables into the dynamic hidden state of a recurrent neural network (RNN) by combining elements of the variational autoencoder. We argue that through the use of high-level latent random variables, the variational RNN (VRNN)1 can model the kind of variability observed in highly structured sequential data such as natural speech. We empirically evaluate the proposed model against related sequential models on four speech datasets and one handwriting dataset. Our results show the important roles that latent random variables can play in the RNN dynamic hidden state.