@inproceedings{xin-etal-2022-zero,
title = "Zero-Shot Dense Retrieval with Momentum Adversarial Domain Invariant Representations",
author = "Xin, Ji and
Xiong, Chenyan and
Srinivasan, Ashwin and
Sharma, Ankita and
Jose, Damien and
Bennett, Paul",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2022",
month = may,
year = "2022",
address = "Dublin, Ireland",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2022.findings-acl.316",
doi = "10.18653/v1/2022.findings-acl.316",
pages = "4008--4020",
abstract = "Dense retrieval (DR) methods conduct text retrieval by first encoding texts in the embedding space and then matching them by nearest neighbor search. This requires strong locality properties from the representation space, e.g., close allocations of each small group of relevant texts, which are hard to generalize to domains without sufficient training data. In this paper, we aim to improve the generalization ability of DR models from source training domains with rich supervision signals to target domains without any relevance label, in the zero-shot setting. To achieve that, we propose Momentum adversarial Domain Invariant Representation learning (MoDIR), which introduces a momentum method to train a domain classifier that distinguishes source versus target domains, and then adversarially updates the DR encoder to learn domain invariant representations. Our experiments show that MoDIR robustly outperforms its baselines on 10+ ranking datasets collected in the BEIR benchmark in the zero-shot setup, with more than 10{\%} relative gains on datasets with enough sensitivity for DR models{'} evaluation. Source code is available at \url{https://github.com/ji-xin/modir}.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="xin-etal-2022-zero">
<titleInfo>
<title>Zero-Shot Dense Retrieval with Momentum Adversarial Domain Invariant Representations</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ji</namePart>
<namePart type="family">Xin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chenyan</namePart>
<namePart type="family">Xiong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ashwin</namePart>
<namePart type="family">Srinivasan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ankita</namePart>
<namePart type="family">Sharma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Damien</namePart>
<namePart type="family">Jose</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Paul</namePart>
<namePart type="family">Bennett</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2022-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2022</title>
</titleInfo>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Dublin, Ireland</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Dense retrieval (DR) methods conduct text retrieval by first encoding texts in the embedding space and then matching them by nearest neighbor search. This requires strong locality properties from the representation space, e.g., close allocations of each small group of relevant texts, which are hard to generalize to domains without sufficient training data. In this paper, we aim to improve the generalization ability of DR models from source training domains with rich supervision signals to target domains without any relevance label, in the zero-shot setting. To achieve that, we propose Momentum adversarial Domain Invariant Representation learning (MoDIR), which introduces a momentum method to train a domain classifier that distinguishes source versus target domains, and then adversarially updates the DR encoder to learn domain invariant representations. Our experiments show that MoDIR robustly outperforms its baselines on 10+ ranking datasets collected in the BEIR benchmark in the zero-shot setup, with more than 10% relative gains on datasets with enough sensitivity for DR models’ evaluation. Source code is available at https://github.com/ji-xin/modir.</abstract>
<identifier type="citekey">xin-etal-2022-zero</identifier>
<identifier type="doi">10.18653/v1/2022.findings-acl.316</identifier>
<location>
<url>https://aclanthology.org/2022.findings-acl.316</url>
</location>
<part>
<date>2022-05</date>
<extent unit="page">
<start>4008</start>
<end>4020</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Zero-Shot Dense Retrieval with Momentum Adversarial Domain Invariant Representations
%A Xin, Ji
%A Xiong, Chenyan
%A Srinivasan, Ashwin
%A Sharma, Ankita
%A Jose, Damien
%A Bennett, Paul
%S Findings of the Association for Computational Linguistics: ACL 2022
%D 2022
%8 May
%I Association for Computational Linguistics
%C Dublin, Ireland
%F xin-etal-2022-zero
%X Dense retrieval (DR) methods conduct text retrieval by first encoding texts in the embedding space and then matching them by nearest neighbor search. This requires strong locality properties from the representation space, e.g., close allocations of each small group of relevant texts, which are hard to generalize to domains without sufficient training data. In this paper, we aim to improve the generalization ability of DR models from source training domains with rich supervision signals to target domains without any relevance label, in the zero-shot setting. To achieve that, we propose Momentum adversarial Domain Invariant Representation learning (MoDIR), which introduces a momentum method to train a domain classifier that distinguishes source versus target domains, and then adversarially updates the DR encoder to learn domain invariant representations. Our experiments show that MoDIR robustly outperforms its baselines on 10+ ranking datasets collected in the BEIR benchmark in the zero-shot setup, with more than 10% relative gains on datasets with enough sensitivity for DR models’ evaluation. Source code is available at https://github.com/ji-xin/modir.
%R 10.18653/v1/2022.findings-acl.316
%U https://aclanthology.org/2022.findings-acl.316
%U https://doi.org/10.18653/v1/2022.findings-acl.316
%P 4008-4020
Markdown (Informal)
[Zero-Shot Dense Retrieval with Momentum Adversarial Domain Invariant Representations](https://aclanthology.org/2022.findings-acl.316) (Xin et al., Findings 2022)
ACL