@inproceedings{jooste-etal-2022-knowledge,
title = "Knowledge Distillation for Sustainable Neural Machine Translation",
author = "Jooste, Wandri and
Way, Andy and
Haque, Rejwanul and
Superbo, Riccardo",
editor = "Campbell, Janice and
Larocca, Stephen and
Marciano, Jay and
Savenkov, Konstantin and
Yanishevsky, Alex",
booktitle = "Proceedings of the 15th Biennial Conference of the Association for Machine Translation in the Americas (Volume 2: Users and Providers Track and Government Track)",
month = sep,
year = "2022",
address = "Orlando, USA",
publisher = "Association for Machine Translation in the Americas",
url = "https://aclanthology.org/2022.amta-upg.16",
pages = "221--230",
abstract = "Knowledge distillation (KD) can be used to reduce model size and training time, without significant loss in performance. However, the process of distilling knowledge requires translation of sizeable data sets, and the translation is usually performed using large cumbersome models (teacher models). Producing such translations for KD is expensive in terms of both time and cost, which is a significant concern for translation service providers. On top of that, this process can be the cause of higher carbon footprints. In this work, we tested different variants of a teacher model for KD, tracked the power consumption of the GPUs used during translation, recorded overall translation time, estimated translation cost, and measured the accuracy of the student models. The findings of our investigation demonstrate to the translation industry a cost-effective, high-quality alternative to the standard KD training methods.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="jooste-etal-2022-knowledge">
<titleInfo>
<title>Knowledge Distillation for Sustainable Neural Machine Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wandri</namePart>
<namePart type="family">Jooste</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andy</namePart>
<namePart type="family">Way</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rejwanul</namePart>
<namePart type="family">Haque</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Riccardo</namePart>
<namePart type="family">Superbo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2022-09</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 15th Biennial Conference of the Association for Machine Translation in the Americas (Volume 2: Users and Providers Track and Government Track)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Janice</namePart>
<namePart type="family">Campbell</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Stephen</namePart>
<namePart type="family">Larocca</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jay</namePart>
<namePart type="family">Marciano</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Konstantin</namePart>
<namePart type="family">Savenkov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alex</namePart>
<namePart type="family">Yanishevsky</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Machine Translation in the Americas</publisher>
<place>
<placeTerm type="text">Orlando, USA</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Knowledge distillation (KD) can be used to reduce model size and training time, without significant loss in performance. However, the process of distilling knowledge requires translation of sizeable data sets, and the translation is usually performed using large cumbersome models (teacher models). Producing such translations for KD is expensive in terms of both time and cost, which is a significant concern for translation service providers. On top of that, this process can be the cause of higher carbon footprints. In this work, we tested different variants of a teacher model for KD, tracked the power consumption of the GPUs used during translation, recorded overall translation time, estimated translation cost, and measured the accuracy of the student models. The findings of our investigation demonstrate to the translation industry a cost-effective, high-quality alternative to the standard KD training methods.</abstract>
<identifier type="citekey">jooste-etal-2022-knowledge</identifier>
<location>
<url>https://aclanthology.org/2022.amta-upg.16</url>
</location>
<part>
<date>2022-09</date>
<extent unit="page">
<start>221</start>
<end>230</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Knowledge Distillation for Sustainable Neural Machine Translation
%A Jooste, Wandri
%A Way, Andy
%A Haque, Rejwanul
%A Superbo, Riccardo
%Y Campbell, Janice
%Y Larocca, Stephen
%Y Marciano, Jay
%Y Savenkov, Konstantin
%Y Yanishevsky, Alex
%S Proceedings of the 15th Biennial Conference of the Association for Machine Translation in the Americas (Volume 2: Users and Providers Track and Government Track)
%D 2022
%8 September
%I Association for Machine Translation in the Americas
%C Orlando, USA
%F jooste-etal-2022-knowledge
%X Knowledge distillation (KD) can be used to reduce model size and training time, without significant loss in performance. However, the process of distilling knowledge requires translation of sizeable data sets, and the translation is usually performed using large cumbersome models (teacher models). Producing such translations for KD is expensive in terms of both time and cost, which is a significant concern for translation service providers. On top of that, this process can be the cause of higher carbon footprints. In this work, we tested different variants of a teacher model for KD, tracked the power consumption of the GPUs used during translation, recorded overall translation time, estimated translation cost, and measured the accuracy of the student models. The findings of our investigation demonstrate to the translation industry a cost-effective, high-quality alternative to the standard KD training methods.
%U https://aclanthology.org/2022.amta-upg.16
%P 221-230
Markdown (Informal)
[Knowledge Distillation for Sustainable Neural Machine Translation](https://aclanthology.org/2022.amta-upg.16) (Jooste et al., AMTA 2022)
ACL
- Wandri Jooste, Andy Way, Rejwanul Haque, and Riccardo Superbo. 2022. Knowledge Distillation for Sustainable Neural Machine Translation. In Proceedings of the 15th Biennial Conference of the Association for Machine Translation in the Americas (Volume 2: Users and Providers Track and Government Track), pages 221–230, Orlando, USA. Association for Machine Translation in the Americas.