@inproceedings{park-etal-2022-consistency,
title = "Consistency Training with Virtual Adversarial Discrete Perturbation",
author = "Park, Jungsoo and
Kim, Gyuwan and
Kang, Jaewoo",
editor = "Carpuat, Marine and
de Marneffe, Marie-Catherine and
Meza Ruiz, Ivan Vladimir",
booktitle = "Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
month = jul,
year = "2022",
address = "Seattle, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2022.naacl-main.414",
doi = "10.18653/v1/2022.naacl-main.414",
pages = "5646--5656",
abstract = "Consistency training regularizes a model by enforcing predictions of original and perturbed inputs to be similar. Previous studies have proposed various augmentation methods for the perturbation but are limited in that they are agnostic to the training model. Thus, the perturbed samples may not aid in regularization due to their ease of classification from the model. In this context, we propose an augmentation method of adding a discrete noise that would incur the highest divergence between predictions. This virtual adversarial discrete noise obtained by replacing a small portion of tokens while keeping original semantics as much as possible efficiently pushes a training model{'}s decision boundary. Experimental results show that our proposed method outperforms other consistency training baselines with text editing, paraphrasing, or a continuous noise on semi-supervised text classification tasks and a robustness benchmark.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="park-etal-2022-consistency">
<titleInfo>
<title>Consistency Training with Virtual Adversarial Discrete Perturbation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jungsoo</namePart>
<namePart type="family">Park</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gyuwan</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jaewoo</namePart>
<namePart type="family">Kang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2022-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies</title>
</titleInfo>
<name type="personal">
<namePart type="given">Marine</namePart>
<namePart type="family">Carpuat</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marie-Catherine</namePart>
<namePart type="family">de Marneffe</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ivan</namePart>
<namePart type="given">Vladimir</namePart>
<namePart type="family">Meza Ruiz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Seattle, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Consistency training regularizes a model by enforcing predictions of original and perturbed inputs to be similar. Previous studies have proposed various augmentation methods for the perturbation but are limited in that they are agnostic to the training model. Thus, the perturbed samples may not aid in regularization due to their ease of classification from the model. In this context, we propose an augmentation method of adding a discrete noise that would incur the highest divergence between predictions. This virtual adversarial discrete noise obtained by replacing a small portion of tokens while keeping original semantics as much as possible efficiently pushes a training model’s decision boundary. Experimental results show that our proposed method outperforms other consistency training baselines with text editing, paraphrasing, or a continuous noise on semi-supervised text classification tasks and a robustness benchmark.</abstract>
<identifier type="citekey">park-etal-2022-consistency</identifier>
<identifier type="doi">10.18653/v1/2022.naacl-main.414</identifier>
<location>
<url>https://aclanthology.org/2022.naacl-main.414</url>
</location>
<part>
<date>2022-07</date>
<extent unit="page">
<start>5646</start>
<end>5656</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Consistency Training with Virtual Adversarial Discrete Perturbation
%A Park, Jungsoo
%A Kim, Gyuwan
%A Kang, Jaewoo
%Y Carpuat, Marine
%Y de Marneffe, Marie-Catherine
%Y Meza Ruiz, Ivan Vladimir
%S Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies
%D 2022
%8 July
%I Association for Computational Linguistics
%C Seattle, United States
%F park-etal-2022-consistency
%X Consistency training regularizes a model by enforcing predictions of original and perturbed inputs to be similar. Previous studies have proposed various augmentation methods for the perturbation but are limited in that they are agnostic to the training model. Thus, the perturbed samples may not aid in regularization due to their ease of classification from the model. In this context, we propose an augmentation method of adding a discrete noise that would incur the highest divergence between predictions. This virtual adversarial discrete noise obtained by replacing a small portion of tokens while keeping original semantics as much as possible efficiently pushes a training model’s decision boundary. Experimental results show that our proposed method outperforms other consistency training baselines with text editing, paraphrasing, or a continuous noise on semi-supervised text classification tasks and a robustness benchmark.
%R 10.18653/v1/2022.naacl-main.414
%U https://aclanthology.org/2022.naacl-main.414
%U https://doi.org/10.18653/v1/2022.naacl-main.414
%P 5646-5656
Markdown (Informal)
[Consistency Training with Virtual Adversarial Discrete Perturbation](https://aclanthology.org/2022.naacl-main.414) (Park et al., NAACL 2022)
ACL
- Jungsoo Park, Gyuwan Kim, and Jaewoo Kang. 2022. Consistency Training with Virtual Adversarial Discrete Perturbation. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5646–5656, Seattle, United States. Association for Computational Linguistics.