@inproceedings{ni-etal-2022-large,
title = "Large Dual Encoders Are Generalizable Retrievers",
author = "Ni, Jianmo and
Qu, Chen and
Lu, Jing and
Dai, Zhuyun and
Hernandez Abrego, Gustavo and
Ma, Ji and
Zhao, Vincent and
Luan, Yi and
Hall, Keith and
Chang, Ming-Wei and
Yang, Yinfei",
editor = "Goldberg, Yoav and
Kozareva, Zornitsa and
Zhang, Yue",
booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing",
month = dec,
year = "2022",
address = "Abu Dhabi, United Arab Emirates",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2022.emnlp-main.669",
doi = "10.18653/v1/2022.emnlp-main.669",
pages = "9844--9855",
abstract = "It has been shown that dual encoders trained on one domain often fail to generalize to other domains for retrieval tasks. One widespread belief is that the bottleneck layer of a dual encoder, where the final score is simply a dot-product between a query vector and a passage vector, is too limited compared to models with fine-grained interactions between the query and the passage. In this paper, we challenge this belief by scaling up the size of the dual encoder model \textit{while keeping the bottleneck layer as a single dot-product with a fixed size.} With multi-stage training, scaling up the model size brings significant improvement on a variety of retrieval tasks, especially for out-of-domain generalization. We further analyze the impact of the bottleneck layer and demonstrate diminishing improvement when scaling up the embedding size. Experimental results show that our dual encoders, \textbf{G}eneralizable \textbf{T}5-based dense \textbf{R}etrievers (GTR), outperform previous sparse and dense retrievers on the BEIR dataset significantly. Most surprisingly, our ablation study finds that GTR is very data efficient, as it only needs 10{\%} of MS Marco supervised data to match the out-of-domain performance of using all supervised data.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ni-etal-2022-large">
<titleInfo>
<title>Large Dual Encoders Are Generalizable Retrievers</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jianmo</namePart>
<namePart type="family">Ni</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chen</namePart>
<namePart type="family">Qu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jing</namePart>
<namePart type="family">Lu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhuyun</namePart>
<namePart type="family">Dai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gustavo</namePart>
<namePart type="family">Hernandez Abrego</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ji</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vincent</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yi</namePart>
<namePart type="family">Luan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Keith</namePart>
<namePart type="family">Hall</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ming-Wei</namePart>
<namePart type="family">Chang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yinfei</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2022-12</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yoav</namePart>
<namePart type="family">Goldberg</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zornitsa</namePart>
<namePart type="family">Kozareva</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Abu Dhabi, United Arab Emirates</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>It has been shown that dual encoders trained on one domain often fail to generalize to other domains for retrieval tasks. One widespread belief is that the bottleneck layer of a dual encoder, where the final score is simply a dot-product between a query vector and a passage vector, is too limited compared to models with fine-grained interactions between the query and the passage. In this paper, we challenge this belief by scaling up the size of the dual encoder model while keeping the bottleneck layer as a single dot-product with a fixed size. With multi-stage training, scaling up the model size brings significant improvement on a variety of retrieval tasks, especially for out-of-domain generalization. We further analyze the impact of the bottleneck layer and demonstrate diminishing improvement when scaling up the embedding size. Experimental results show that our dual encoders, Generalizable T5-based dense Retrievers (GTR), outperform previous sparse and dense retrievers on the BEIR dataset significantly. Most surprisingly, our ablation study finds that GTR is very data efficient, as it only needs 10% of MS Marco supervised data to match the out-of-domain performance of using all supervised data.</abstract>
<identifier type="citekey">ni-etal-2022-large</identifier>
<identifier type="doi">10.18653/v1/2022.emnlp-main.669</identifier>
<location>
<url>https://aclanthology.org/2022.emnlp-main.669</url>
</location>
<part>
<date>2022-12</date>
<extent unit="page">
<start>9844</start>
<end>9855</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Large Dual Encoders Are Generalizable Retrievers
%A Ni, Jianmo
%A Qu, Chen
%A Lu, Jing
%A Dai, Zhuyun
%A Hernandez Abrego, Gustavo
%A Ma, Ji
%A Zhao, Vincent
%A Luan, Yi
%A Hall, Keith
%A Chang, Ming-Wei
%A Yang, Yinfei
%Y Goldberg, Yoav
%Y Kozareva, Zornitsa
%Y Zhang, Yue
%S Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing
%D 2022
%8 December
%I Association for Computational Linguistics
%C Abu Dhabi, United Arab Emirates
%F ni-etal-2022-large
%X It has been shown that dual encoders trained on one domain often fail to generalize to other domains for retrieval tasks. One widespread belief is that the bottleneck layer of a dual encoder, where the final score is simply a dot-product between a query vector and a passage vector, is too limited compared to models with fine-grained interactions between the query and the passage. In this paper, we challenge this belief by scaling up the size of the dual encoder model while keeping the bottleneck layer as a single dot-product with a fixed size. With multi-stage training, scaling up the model size brings significant improvement on a variety of retrieval tasks, especially for out-of-domain generalization. We further analyze the impact of the bottleneck layer and demonstrate diminishing improvement when scaling up the embedding size. Experimental results show that our dual encoders, Generalizable T5-based dense Retrievers (GTR), outperform previous sparse and dense retrievers on the BEIR dataset significantly. Most surprisingly, our ablation study finds that GTR is very data efficient, as it only needs 10% of MS Marco supervised data to match the out-of-domain performance of using all supervised data.
%R 10.18653/v1/2022.emnlp-main.669
%U https://aclanthology.org/2022.emnlp-main.669
%U https://doi.org/10.18653/v1/2022.emnlp-main.669
%P 9844-9855
Markdown (Informal)
[Large Dual Encoders Are Generalizable Retrievers](https://aclanthology.org/2022.emnlp-main.669) (Ni et al., EMNLP 2022)
ACL
- Jianmo Ni, Chen Qu, Jing Lu, Zhuyun Dai, Gustavo Hernandez Abrego, Ji Ma, Vincent Zhao, Yi Luan, Keith Hall, Ming-Wei Chang, and Yinfei Yang. 2022. Large Dual Encoders Are Generalizable Retrievers. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 9844–9855, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics.