Abstract
In this paper we explore the effect of architectural choices on learning a variational autoencoder (VAE) for text generation. In contrast to the previously introduced VAE model for text where both the encoder and decoder are RNNs, we propose a novel hybrid architecture that blends fully feed-forward convolutional and deconvolutional components with a recurrent language model. Our architecture exhibits several attractive properties such as faster run time and convergence, ability to better handle long sequences and, more importantly, it helps to avoid the issue of the VAE collapsing to a deterministic model.
Originalsprache | Englisch |
---|---|
Titel | EMNLP 2017 - Conference on Empirical Methods in Natural Language Processing, Proceedings |
Seitenumfang | 11 |
Herausgeber (Verlag) | Association for Computational Linguistics (ACL) |
Erscheinungsdatum | 09.2017 |
Seiten | 627–637 |
ISBN (Print) | 978-194562683-8 |
DOIs | |
Publikationsstatus | Veröffentlicht - 09.2017 |
Veranstaltung | 2017 Conference on Empirical Methods in Natural Language Processing - Copenhagen, Dänemark Dauer: 09.09.2017 → 11.09.2017 Konferenznummer: 150071 |