J. Wu, L. Li, und W. Wang. (2018)cite arxiv:1804.06035Comment: 11 pages, 3 figures. Accepted to NAACL 2018.
Zusammenfassung
Co-training is a popular semi-supervised learning framework to utilize a
large amount of unlabeled data in addition to a small labeled set. Co-training
methods exploit predicted labels on the unlabeled data and select samples based
on prediction confidence to augment the training. However, the selection of
samples in existing co-training methods is based on a predetermined policy,
which ignores the sampling bias between the unlabeled and the labeled subsets,
and fails to explore the data space. In this paper, we propose a novel method,
Reinforced Co-Training, to select high-quality unlabeled samples to better
co-train on. More specifically, our approach uses Q-learning to learn a data
selection policy with a small labeled dataset, and then exploits this policy to
train the co-training classifiers automatically. Experimental results on
clickbait detection and generic text classification tasks demonstrate that our
proposed method can obtain more accurate text classification results.
%0 Generic
%1 wu2018reinforced
%A Wu, Jiawei
%A Li, Lei
%A Wang, William Yang
%D 2018
%K semisup
%T Reinforced Co-Training
%U http://arxiv.org/abs/1804.06035
%X Co-training is a popular semi-supervised learning framework to utilize a
large amount of unlabeled data in addition to a small labeled set. Co-training
methods exploit predicted labels on the unlabeled data and select samples based
on prediction confidence to augment the training. However, the selection of
samples in existing co-training methods is based on a predetermined policy,
which ignores the sampling bias between the unlabeled and the labeled subsets,
and fails to explore the data space. In this paper, we propose a novel method,
Reinforced Co-Training, to select high-quality unlabeled samples to better
co-train on. More specifically, our approach uses Q-learning to learn a data
selection policy with a small labeled dataset, and then exploits this policy to
train the co-training classifiers automatically. Experimental results on
clickbait detection and generic text classification tasks demonstrate that our
proposed method can obtain more accurate text classification results.
@misc{wu2018reinforced,
abstract = {Co-training is a popular semi-supervised learning framework to utilize a
large amount of unlabeled data in addition to a small labeled set. Co-training
methods exploit predicted labels on the unlabeled data and select samples based
on prediction confidence to augment the training. However, the selection of
samples in existing co-training methods is based on a predetermined policy,
which ignores the sampling bias between the unlabeled and the labeled subsets,
and fails to explore the data space. In this paper, we propose a novel method,
Reinforced Co-Training, to select high-quality unlabeled samples to better
co-train on. More specifically, our approach uses Q-learning to learn a data
selection policy with a small labeled dataset, and then exploits this policy to
train the co-training classifiers automatically. Experimental results on
clickbait detection and generic text classification tasks demonstrate that our
proposed method can obtain more accurate text classification results.},
added-at = {2018-05-07T10:08:46.000+0200},
author = {Wu, Jiawei and Li, Lei and Wang, William Yang},
biburl = {https://www.bibsonomy.org/bibtex/228a080e83de35b95786c133b897666e1/topel},
interhash = {bbac65bd7647923fa4084d473f989a00},
intrahash = {28a080e83de35b95786c133b897666e1},
keywords = {semisup},
note = {cite arxiv:1804.06035Comment: 11 pages, 3 figures. Accepted to NAACL 2018},
timestamp = {2018-05-07T10:09:16.000+0200},
title = {Reinforced Co-Training},
url = {http://arxiv.org/abs/1804.06035},
year = 2018
}