@inproceedings{reinhardt-etal-2024-improving,
title = "Improving Vision-Language Cross-Lingual Transfer with Scheduled Unfreezing",
author = "Reinhardt, Max and
Geigle, Gregor and
Timofte, Radu and
Glava{\v{s}}, Goran",
editor = "Gu, Jing and
Fu, Tsu-Jui (Ray) and
Hudson, Drew and
Celikyilmaz, Asli and
Wang, William",
booktitle = "Proceedings of the 3rd Workshop on Advances in Language and Vision Research (ALVR)",
month = aug,
year = "2024",
address = "Bangkok, Thailand",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.alvr-1.13",
pages = "155--166",
abstract = "Large-scale pretraining of vision-language (VL) models brought dramatic improvements across numerous tasks, from visual question-answering to cross-modal retrieval but these gains are mostly limited to English. Massively multilingual VL encoder models (mVLMs) hold promise for other languages: after fine-tuning on only English task data, they can perform the task in other languages in what is termed zero-shot cross-lingual transfer (ZS-XLT). Still, ZS-XLT sees a large performance gap to English, especially for low-resource languages. In this work, we reduce this gap with a fine-tuning strategy known as \textit{Scheduled Unfreezing} (SUF): instead of updating all parameters from the start, we begin with the top layer(s) of the vision-language encoder and gradually unfreeze (i.e., update) its layers top to bottom. SUF forces reliance on encoder{'}s representations from higher layers: the fact that in multilingual models these representations encode higher-level semantics rather than low-level language-specific idiosyncrasies, we hypothesize, should render SUF beneficial for ZS-XLT. Experiments with two mVLMs (UC2 {\&} CCLM) on three downstream tasks (xGQA, XVNLI, xFlickrCo) show that SUF brings consistent gains in ZS-XLT, especially for visual Q{\&}A (xGQA) by up to 10 points.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="reinhardt-etal-2024-improving">
<titleInfo>
<title>Improving Vision-Language Cross-Lingual Transfer with Scheduled Unfreezing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Max</namePart>
<namePart type="family">Reinhardt</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gregor</namePart>
<namePart type="family">Geigle</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Radu</namePart>
<namePart type="family">Timofte</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Goran</namePart>
<namePart type="family">Glavaš</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-08</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 3rd Workshop on Advances in Language and Vision Research (ALVR)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jing</namePart>
<namePart type="family">Gu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tsu-Jui</namePart>
<namePart type="given">(Ray)</namePart>
<namePart type="family">Fu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Drew</namePart>
<namePart type="family">Hudson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Asli</namePart>
<namePart type="family">Celikyilmaz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">William</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Bangkok, Thailand</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Large-scale pretraining of vision-language (VL) models brought dramatic improvements across numerous tasks, from visual question-answering to cross-modal retrieval but these gains are mostly limited to English. Massively multilingual VL encoder models (mVLMs) hold promise for other languages: after fine-tuning on only English task data, they can perform the task in other languages in what is termed zero-shot cross-lingual transfer (ZS-XLT). Still, ZS-XLT sees a large performance gap to English, especially for low-resource languages. In this work, we reduce this gap with a fine-tuning strategy known as Scheduled Unfreezing (SUF): instead of updating all parameters from the start, we begin with the top layer(s) of the vision-language encoder and gradually unfreeze (i.e., update) its layers top to bottom. SUF forces reliance on encoder’s representations from higher layers: the fact that in multilingual models these representations encode higher-level semantics rather than low-level language-specific idiosyncrasies, we hypothesize, should render SUF beneficial for ZS-XLT. Experiments with two mVLMs (UC2 & CCLM) on three downstream tasks (xGQA, XVNLI, xFlickrCo) show that SUF brings consistent gains in ZS-XLT, especially for visual Q&A (xGQA) by up to 10 points.</abstract>
<identifier type="citekey">reinhardt-etal-2024-improving</identifier>
<location>
<url>https://aclanthology.org/2024.alvr-1.13</url>
</location>
<part>
<date>2024-08</date>
<extent unit="page">
<start>155</start>
<end>166</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Improving Vision-Language Cross-Lingual Transfer with Scheduled Unfreezing
%A Reinhardt, Max
%A Geigle, Gregor
%A Timofte, Radu
%A Glavaš, Goran
%Y Gu, Jing
%Y Fu, Tsu-Jui (Ray)
%Y Hudson, Drew
%Y Celikyilmaz, Asli
%Y Wang, William
%S Proceedings of the 3rd Workshop on Advances in Language and Vision Research (ALVR)
%D 2024
%8 August
%I Association for Computational Linguistics
%C Bangkok, Thailand
%F reinhardt-etal-2024-improving
%X Large-scale pretraining of vision-language (VL) models brought dramatic improvements across numerous tasks, from visual question-answering to cross-modal retrieval but these gains are mostly limited to English. Massively multilingual VL encoder models (mVLMs) hold promise for other languages: after fine-tuning on only English task data, they can perform the task in other languages in what is termed zero-shot cross-lingual transfer (ZS-XLT). Still, ZS-XLT sees a large performance gap to English, especially for low-resource languages. In this work, we reduce this gap with a fine-tuning strategy known as Scheduled Unfreezing (SUF): instead of updating all parameters from the start, we begin with the top layer(s) of the vision-language encoder and gradually unfreeze (i.e., update) its layers top to bottom. SUF forces reliance on encoder’s representations from higher layers: the fact that in multilingual models these representations encode higher-level semantics rather than low-level language-specific idiosyncrasies, we hypothesize, should render SUF beneficial for ZS-XLT. Experiments with two mVLMs (UC2 & CCLM) on three downstream tasks (xGQA, XVNLI, xFlickrCo) show that SUF brings consistent gains in ZS-XLT, especially for visual Q&A (xGQA) by up to 10 points.
%U https://aclanthology.org/2024.alvr-1.13
%P 155-166
Markdown (Informal)
[Improving Vision-Language Cross-Lingual Transfer with Scheduled Unfreezing](https://aclanthology.org/2024.alvr-1.13) (Reinhardt et al., ALVR-WS 2024)
ACL