Iterated dilated convolutional neural networks for word segmentation

The latest development of neural word segmentation is governed by bi-directional Long Short-Term Memory Networks (Bi-LSTMs) that utilize Recurrent Neural Networks (RNNs) as standard sequence tagging models, resulting in expressive and accurate performance on large-scale dataset. However, RNNs are no...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Neural network world Jg. 30; H. 5; S. 333
Hauptverfasser: He. H., Yang, X, Wu L., Wang, G
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Prague Czech Technical University in Prague, Faculty of Transportation Sciences 01.01.2020
Schlagworte:
Accuracy
Artificial neural networks
Datasets
Graphics processing units
Neural networks
Recurrent neural networks
Segmentation
Source code
ISSN:1210-0552, 2336-4335
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The latest development of neural word segmentation is governed by bi-directional Long Short-Term Memory Networks (Bi-LSTMs) that utilize Recurrent Neural Networks (RNNs) as standard sequence tagging models, resulting in expressive and accurate performance on large-scale dataset. However, RNNs are not adapted to fully exploit the parallelism capability of Graphics Processing Unit (GPU), limiting their computational efficiency in both learning and inferring phases. This paper proposes a novel approach adopting Iterated Dilated Convolutional Neural Networks (ID-CNNs) to supersede Bi-LSTMs for faster computation while retaining accuracy. Our implementation has achieved state-of-the-art result on SIGHAN Bakeoff 2005 datasets. Extensive experiments showed that our approach with ID-CNNs enables 3x training time speedups with no accuracy loss, achieving better accuracy compared to the prevailing Bi-LSTMs. Source code and corpora of this paper have been made publicly available on GitHub.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1210-0552
2336-4335
DOI:10.14311/nnw.2020.30.022