@inproceedings{sar-shalom-schwartz-2023-curating,
title = "Curating Datasets for Better Performance with Example Training Dynamics",
author = "Sar-Shalom, Aviad and
Schwartz, Roy",
editor = "Rogers, Anna and
Boyd-Graber, Jordan and
Okazaki, Naoaki",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.findings-acl.674",
doi = "10.18653/v1/2023.findings-acl.674",
pages = "10597--10608",
abstract = "The landscape of NLP research is dominated by large-scale models training on colossal datasets, relying on data quantity rather than quality. As an alternative to this landscape, we propose a method for weighing the relative importance of examples in a dataset based on their Example Training dynamics (swayamdipta et al., 2020) {---} a set of metrics computed during training. We propose a new way of computing the ETD of a dataset, and show that they can be used to improve performance in both in-distribution and out-of-distribution testing. We show that ETD can be transferable, i.e., they can be computed once and used for training different models, effectively reducing their computation cost. Finally, we suggest an active learning approach for computing ETD during training rather than as a preprocessing step {---} an approach that is not as effective, but dramatically reduces the extra computational costs.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="sar-shalom-schwartz-2023-curating">
<titleInfo>
<title>Curating Datasets for Better Performance with Example Training Dynamics</title>
</titleInfo>
<name type="personal">
<namePart type="given">Aviad</namePart>
<namePart type="family">Sar-Shalom</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Roy</namePart>
<namePart type="family">Schwartz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2023-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2023</title>
</titleInfo>
<name type="personal">
<namePart type="given">Anna</namePart>
<namePart type="family">Rogers</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jordan</namePart>
<namePart type="family">Boyd-Graber</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naoaki</namePart>
<namePart type="family">Okazaki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Toronto, Canada</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>The landscape of NLP research is dominated by large-scale models training on colossal datasets, relying on data quantity rather than quality. As an alternative to this landscape, we propose a method for weighing the relative importance of examples in a dataset based on their Example Training dynamics (swayamdipta et al., 2020) — a set of metrics computed during training. We propose a new way of computing the ETD of a dataset, and show that they can be used to improve performance in both in-distribution and out-of-distribution testing. We show that ETD can be transferable, i.e., they can be computed once and used for training different models, effectively reducing their computation cost. Finally, we suggest an active learning approach for computing ETD during training rather than as a preprocessing step — an approach that is not as effective, but dramatically reduces the extra computational costs.</abstract>
<identifier type="citekey">sar-shalom-schwartz-2023-curating</identifier>
<identifier type="doi">10.18653/v1/2023.findings-acl.674</identifier>
<location>
<url>https://aclanthology.org/2023.findings-acl.674</url>
</location>
<part>
<date>2023-07</date>
<extent unit="page">
<start>10597</start>
<end>10608</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Curating Datasets for Better Performance with Example Training Dynamics
%A Sar-Shalom, Aviad
%A Schwartz, Roy
%Y Rogers, Anna
%Y Boyd-Graber, Jordan
%Y Okazaki, Naoaki
%S Findings of the Association for Computational Linguistics: ACL 2023
%D 2023
%8 July
%I Association for Computational Linguistics
%C Toronto, Canada
%F sar-shalom-schwartz-2023-curating
%X The landscape of NLP research is dominated by large-scale models training on colossal datasets, relying on data quantity rather than quality. As an alternative to this landscape, we propose a method for weighing the relative importance of examples in a dataset based on their Example Training dynamics (swayamdipta et al., 2020) — a set of metrics computed during training. We propose a new way of computing the ETD of a dataset, and show that they can be used to improve performance in both in-distribution and out-of-distribution testing. We show that ETD can be transferable, i.e., they can be computed once and used for training different models, effectively reducing their computation cost. Finally, we suggest an active learning approach for computing ETD during training rather than as a preprocessing step — an approach that is not as effective, but dramatically reduces the extra computational costs.
%R 10.18653/v1/2023.findings-acl.674
%U https://aclanthology.org/2023.findings-acl.674
%U https://doi.org/10.18653/v1/2023.findings-acl.674
%P 10597-10608
Markdown (Informal)
[Curating Datasets for Better Performance with Example Training Dynamics](https://aclanthology.org/2023.findings-acl.674) (Sar-Shalom & Schwartz, Findings 2023)
ACL