Learning deep autoregressive models for hierarchical data

TL;DR

This paper introduces a deep autoregressive model with hierarchical variational autoencoders and downsampling, designed for hierarchical data, demonstrating state-of-the-art results on speech and handwritten text sequences.

Contribution

It presents a novel combination of autoregressive modeling with hierarchical VAEs and downsampling for improved efficiency and performance on sequential data.

Findings

Achieved state-of-the-art performance on speech data

Performed well on handwritten text sequences

Enhanced computational efficiency through downsampling

Abstract

We propose a model for hierarchical structured data as an extension to the stochastic temporal convolutional network. The proposed model combines an autoregressive model with a hierarchical variational autoencoder and downsampling to achieve superior computational complexity. We evaluate the proposed model on two different types of sequential data: speech and handwritten text. The results are promising with the proposed model achieving state-of-the-art performance.