@inproceedings{tran-etal-2026-pga,
title = "{PGA}-{KD}: Principal Geometry Alignment and Semantic Consistency for {VLM}2{V}ec Distillation",
author = "Tran, The-Anh and
Nguyen, Thanh Xuan and
Nguyen, Duc Anh and
Sang, Dinh Viet and
Van, Linh Ngo and
Nguyen, Thien Huu",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.257/",
pages = "5827--5847",
ISBN = "979-8-89176-459-0",
abstract = "Adapting large vision{--}language models for embedding tasks (VLM2Vec) is standard practice, but distilling them into compact architectures remains challenging due to capacity mismatches in tokenization, dimensionality, and multimodal representations. To address this, we propose a robust distillation framework based on two complementary mechanisms: Principal Geometry Alignment (PGA) to stabilize training by aligning principal geometric components, and a Mutual Information Maximization objective to enforce global semantic consistency across intra- and inter-modal pairs. By jointly constraining geometric structure and semantic dependency, our method effectively mitigates model capacity discrepancies. Experimental results on the Massive Multimodal Embedding Benchmark (MMEB) across all four meta-tasks{---}under both In-Distribution and Out-of-Distribution settings{---}demonstrate that our approach consistently outperforms state-of-the-art distillation methods. Our source code is available at https://github.com/theanh-ktmt/PGA-KD."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="tran-etal-2026-pga">
<titleInfo>
<title>PGA-KD: Principal Geometry Alignment and Semantic Consistency for VLM2Vec Distillation</title>
</titleInfo>
<name type="personal">
<namePart type="given">The-Anh</namePart>
<namePart type="family">Tran</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thanh</namePart>
<namePart type="given">Xuan</namePart>
<namePart type="family">Nguyen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Duc</namePart>
<namePart type="given">Anh</namePart>
<namePart type="family">Nguyen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dinh</namePart>
<namePart type="given">Viet</namePart>
<namePart type="family">Sang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Linh</namePart>
<namePart type="given">Ngo</namePart>
<namePart type="family">Van</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thien</namePart>
<namePart type="given">Huu</namePart>
<namePart type="family">Nguyen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Adapting large vision–language models for embedding tasks (VLM2Vec) is standard practice, but distilling them into compact architectures remains challenging due to capacity mismatches in tokenization, dimensionality, and multimodal representations. To address this, we propose a robust distillation framework based on two complementary mechanisms: Principal Geometry Alignment (PGA) to stabilize training by aligning principal geometric components, and a Mutual Information Maximization objective to enforce global semantic consistency across intra- and inter-modal pairs. By jointly constraining geometric structure and semantic dependency, our method effectively mitigates model capacity discrepancies. Experimental results on the Massive Multimodal Embedding Benchmark (MMEB) across all four meta-tasks—under both In-Distribution and Out-of-Distribution settings—demonstrate that our approach consistently outperforms state-of-the-art distillation methods. Our source code is available at https://github.com/theanh-ktmt/PGA-KD.</abstract>
<identifier type="citekey">tran-etal-2026-pga</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.257/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>5827</start>
<end>5847</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T PGA-KD: Principal Geometry Alignment and Semantic Consistency for VLM2Vec Distillation
%A Tran, The-Anh
%A Nguyen, Thanh Xuan
%A Nguyen, Duc Anh
%A Sang, Dinh Viet
%A Van, Linh Ngo
%A Nguyen, Thien Huu
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F tran-etal-2026-pga
%X Adapting large vision–language models for embedding tasks (VLM2Vec) is standard practice, but distilling them into compact architectures remains challenging due to capacity mismatches in tokenization, dimensionality, and multimodal representations. To address this, we propose a robust distillation framework based on two complementary mechanisms: Principal Geometry Alignment (PGA) to stabilize training by aligning principal geometric components, and a Mutual Information Maximization objective to enforce global semantic consistency across intra- and inter-modal pairs. By jointly constraining geometric structure and semantic dependency, our method effectively mitigates model capacity discrepancies. Experimental results on the Massive Multimodal Embedding Benchmark (MMEB) across all four meta-tasks—under both In-Distribution and Out-of-Distribution settings—demonstrate that our approach consistently outperforms state-of-the-art distillation methods. Our source code is available at https://github.com/theanh-ktmt/PGA-KD.
%U https://aclanthology.org/2026.emnlp-main.257/
%P 5827-5847
Markdown (Informal)
[PGA-KD: Principal Geometry Alignment and Semantic Consistency for VLM2Vec Distillation](https://aclanthology.org/2026.emnlp-main.257/) (Tran et al., EMNLP 2026)
ACL