@inproceedings{chang-lu-2021-rethinking-intermediate,
title = "Rethinking Why Intermediate-Task Fine-Tuning Works",
author = "Chang, Ting-Yun and
Lu, Chi-Jen",
editor = "Moens, Marie-Francine and
Huang, Xuanjing and
Specia, Lucia and
Yih, Scott Wen-tau",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2021",
month = nov,
year = "2021",
address = "Punta Cana, Dominican Republic",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.findings-emnlp.61",
doi = "10.18653/v1/2021.findings-emnlp.61",
pages = "706--713",
abstract = "Supplementary Training on Intermediate Labeled-data Tasks (STILT) is a widely applied technique, which first fine-tunes the pretrained language models on an intermediate task before on the target task of interest. While STILT is able to further improve the performance of pretrained language models, it is still unclear why and when it works. Previous research shows that those intermediate tasks involving complex inference, such as commonsense reasoning, work especially well for RoBERTa-large. In this paper, we discover that the improvement from an intermediate task could be orthogonal to it containing reasoning or other complex skills {---} a simple real-fake discrimination task synthesized by GPT2 can benefit diverse target tasks. We conduct extensive experiments to study the impact of different factors on STILT. These findings suggest rethinking the role of intermediate fine-tuning in the STILT pipeline.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chang-lu-2021-rethinking-intermediate">
<titleInfo>
<title>Rethinking Why Intermediate-Task Fine-Tuning Works</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ting-Yun</namePart>
<namePart type="family">Chang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chi-Jen</namePart>
<namePart type="family">Lu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2021-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2021</title>
</titleInfo>
<name type="personal">
<namePart type="given">Marie-Francine</namePart>
<namePart type="family">Moens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xuanjing</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lucia</namePart>
<namePart type="family">Specia</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Scott</namePart>
<namePart type="given">Wen-tau</namePart>
<namePart type="family">Yih</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Punta Cana, Dominican Republic</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Supplementary Training on Intermediate Labeled-data Tasks (STILT) is a widely applied technique, which first fine-tunes the pretrained language models on an intermediate task before on the target task of interest. While STILT is able to further improve the performance of pretrained language models, it is still unclear why and when it works. Previous research shows that those intermediate tasks involving complex inference, such as commonsense reasoning, work especially well for RoBERTa-large. In this paper, we discover that the improvement from an intermediate task could be orthogonal to it containing reasoning or other complex skills — a simple real-fake discrimination task synthesized by GPT2 can benefit diverse target tasks. We conduct extensive experiments to study the impact of different factors on STILT. These findings suggest rethinking the role of intermediate fine-tuning in the STILT pipeline.</abstract>
<identifier type="citekey">chang-lu-2021-rethinking-intermediate</identifier>
<identifier type="doi">10.18653/v1/2021.findings-emnlp.61</identifier>
<location>
<url>https://aclanthology.org/2021.findings-emnlp.61</url>
</location>
<part>
<date>2021-11</date>
<extent unit="page">
<start>706</start>
<end>713</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Rethinking Why Intermediate-Task Fine-Tuning Works
%A Chang, Ting-Yun
%A Lu, Chi-Jen
%Y Moens, Marie-Francine
%Y Huang, Xuanjing
%Y Specia, Lucia
%Y Yih, Scott Wen-tau
%S Findings of the Association for Computational Linguistics: EMNLP 2021
%D 2021
%8 November
%I Association for Computational Linguistics
%C Punta Cana, Dominican Republic
%F chang-lu-2021-rethinking-intermediate
%X Supplementary Training on Intermediate Labeled-data Tasks (STILT) is a widely applied technique, which first fine-tunes the pretrained language models on an intermediate task before on the target task of interest. While STILT is able to further improve the performance of pretrained language models, it is still unclear why and when it works. Previous research shows that those intermediate tasks involving complex inference, such as commonsense reasoning, work especially well for RoBERTa-large. In this paper, we discover that the improvement from an intermediate task could be orthogonal to it containing reasoning or other complex skills — a simple real-fake discrimination task synthesized by GPT2 can benefit diverse target tasks. We conduct extensive experiments to study the impact of different factors on STILT. These findings suggest rethinking the role of intermediate fine-tuning in the STILT pipeline.
%R 10.18653/v1/2021.findings-emnlp.61
%U https://aclanthology.org/2021.findings-emnlp.61
%U https://doi.org/10.18653/v1/2021.findings-emnlp.61
%P 706-713
Markdown (Informal)
[Rethinking Why Intermediate-Task Fine-Tuning Works](https://aclanthology.org/2021.findings-emnlp.61) (Chang & Lu, Findings 2021)
ACL
- Ting-Yun Chang and Chi-Jen Lu. 2021. Rethinking Why Intermediate-Task Fine-Tuning Works. In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 706–713, Punta Cana, Dominican Republic. Association for Computational Linguistics.