The dominant approach for many NLP tasks are recurrent neural networks, in
particular LSTMs, and convolutional neural networks. However, these
architectures are rather shallow in comparison to the deep convolutional
networks which are very successful in computer vision. We present a new
architecture for text processing which operates directly on the character level
and uses only small convolutions and pooling operations. We are able to show
that the performance of this model increases with the depth: using up to 29
convolutional layers, we report significant improvements over the
state-of-the-art on several public text classification tasks. To the best of
our knowledge, this is the first time that very deep convolutional nets have
been applied to NLP.
Description
Very Deep Convolutional Networks for Natural Language Processing
%0 Generic
%1 conneau2016convolutional
%A Conneau, Alexis
%A Schwenk, Holger
%A Barrault, Loïc
%A Lecun, Yann
%D 2016
%K character-level cnn gpu kallimachos nlp
%T Very Deep Convolutional Networks for Natural Language Processing
%U http://arxiv.org/abs/1606.01781
%X The dominant approach for many NLP tasks are recurrent neural networks, in
particular LSTMs, and convolutional neural networks. However, these
architectures are rather shallow in comparison to the deep convolutional
networks which are very successful in computer vision. We present a new
architecture for text processing which operates directly on the character level
and uses only small convolutions and pooling operations. We are able to show
that the performance of this model increases with the depth: using up to 29
convolutional layers, we report significant improvements over the
state-of-the-art on several public text classification tasks. To the best of
our knowledge, this is the first time that very deep convolutional nets have
been applied to NLP.
@misc{conneau2016convolutional,
abstract = {The dominant approach for many NLP tasks are recurrent neural networks, in
particular LSTMs, and convolutional neural networks. However, these
architectures are rather shallow in comparison to the deep convolutional
networks which are very successful in computer vision. We present a new
architecture for text processing which operates directly on the character level
and uses only small convolutions and pooling operations. We are able to show
that the performance of this model increases with the depth: using up to 29
convolutional layers, we report significant improvements over the
state-of-the-art on several public text classification tasks. To the best of
our knowledge, this is the first time that very deep convolutional nets have
been applied to NLP.},
added-at = {2017-01-18T15:40:36.000+0100},
author = {Conneau, Alexis and Schwenk, Holger and Barrault, Loïc and Lecun, Yann},
biburl = {https://www.bibsonomy.org/bibtex/2674f7270e51fe196bc83f9d2ff1e9563/albinzehe},
description = {Very Deep Convolutional Networks for Natural Language Processing},
interhash = {cd8654a7ab8b4c8c2c124c18b96ff5b1},
intrahash = {674f7270e51fe196bc83f9d2ff1e9563},
keywords = {character-level cnn gpu kallimachos nlp},
note = {cite arxiv:1606.01781},
timestamp = {2017-01-18T15:42:50.000+0100},
title = {Very Deep Convolutional Networks for Natural Language Processing},
url = {http://arxiv.org/abs/1606.01781},
year = 2016
}