@inproceedings{do-etal-2026-vlm2vec,
title = "{M}-{VLM}2{V}ec: Structure-Preserving Matryoshka Representation Learning for Vision-Language Embeddings",
author = "Do, Minh Khoi Nguyen and
Dang, Nguyen Hong and
Le, Danh Vinh and
Nguyen, Minh Quoc Nhat and
Van, Linh Ngo and
Nguyen, Duc Anh and
Le, Trung",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.308/",
pages = "6515--6537",
ISBN = "979-8-89176-462-0",
abstract = "Deploying high-dimensional Vision-Language Model (VLM) embeddings is computationally prohibitive. While Matryoshka Representation Learning (MRL) allows elastic truncation, standard MRL optimizes dimensional hierarchies semi-independently. This lack of cross-scale structural constraints creates a severe topological disconnect causing representational collapse at lower dimensions and introduces optimization trade-offs that compromise high-dimensional capacity. To solve this, we propose **M-VLM2Vec**, a framework that shifts MRL from a compromised hierarchy to a synergistic geometric structure. Without additional supervision, we integrate two novel constraints into the multimodal contrastive objective: (1) Laplacian spectral alignment to preserve the global manifold topology across scales, and (2) Stiefel-constrained orthogonal projection to bidirectionally align local contrastive landscapes. By allowing nested spaces to mutually regularize, M-VLM2Vec eliminates competing gradients, preventing ultra-low-dimensional collapse while yielding striking, consistent performance gains across all embedding scales on the Multimodal Embedding Benchmark (MMEB)."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="do-etal-2026-vlm2vec">
<titleInfo>
<title>M-VLM2Vec: Structure-Preserving Matryoshka Representation Learning for Vision-Language Embeddings</title>
</titleInfo>
<name type="personal">
<namePart type="given">Minh</namePart>
<namePart type="given">Khoi</namePart>
<namePart type="given">Nguyen</namePart>
<namePart type="family">Do</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nguyen</namePart>
<namePart type="given">Hong</namePart>
<namePart type="family">Dang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Danh</namePart>
<namePart type="given">Vinh</namePart>
<namePart type="family">Le</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Minh</namePart>
<namePart type="given">Quoc</namePart>
<namePart type="given">Nhat</namePart>
<namePart type="family">Nguyen</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">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">Trung</namePart>
<namePart type="family">Le</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Deploying high-dimensional Vision-Language Model (VLM) embeddings is computationally prohibitive. While Matryoshka Representation Learning (MRL) allows elastic truncation, standard MRL optimizes dimensional hierarchies semi-independently. This lack of cross-scale structural constraints creates a severe topological disconnect causing representational collapse at lower dimensions and introduces optimization trade-offs that compromise high-dimensional capacity. To solve this, we propose **M-VLM2Vec**, a framework that shifts MRL from a compromised hierarchy to a synergistic geometric structure. Without additional supervision, we integrate two novel constraints into the multimodal contrastive objective: (1) Laplacian spectral alignment to preserve the global manifold topology across scales, and (2) Stiefel-constrained orthogonal projection to bidirectionally align local contrastive landscapes. By allowing nested spaces to mutually regularize, M-VLM2Vec eliminates competing gradients, preventing ultra-low-dimensional collapse while yielding striking, consistent performance gains across all embedding scales on the Multimodal Embedding Benchmark (MMEB).</abstract>
<identifier type="citekey">do-etal-2026-vlm2vec</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.308/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>6515</start>
<end>6537</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T M-VLM2Vec: Structure-Preserving Matryoshka Representation Learning for Vision-Language Embeddings
%A Do, Minh Khoi Nguyen
%A Dang, Nguyen Hong
%A Le, Danh Vinh
%A Nguyen, Minh Quoc Nhat
%A Van, Linh Ngo
%A Nguyen, Duc Anh
%A Le, Trung
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F do-etal-2026-vlm2vec
%X Deploying high-dimensional Vision-Language Model (VLM) embeddings is computationally prohibitive. While Matryoshka Representation Learning (MRL) allows elastic truncation, standard MRL optimizes dimensional hierarchies semi-independently. This lack of cross-scale structural constraints creates a severe topological disconnect causing representational collapse at lower dimensions and introduces optimization trade-offs that compromise high-dimensional capacity. To solve this, we propose **M-VLM2Vec**, a framework that shifts MRL from a compromised hierarchy to a synergistic geometric structure. Without additional supervision, we integrate two novel constraints into the multimodal contrastive objective: (1) Laplacian spectral alignment to preserve the global manifold topology across scales, and (2) Stiefel-constrained orthogonal projection to bidirectionally align local contrastive landscapes. By allowing nested spaces to mutually regularize, M-VLM2Vec eliminates competing gradients, preventing ultra-low-dimensional collapse while yielding striking, consistent performance gains across all embedding scales on the Multimodal Embedding Benchmark (MMEB).
%U https://aclanthology.org/2026.findings-emnlp.308/
%P 6515-6537
Markdown (Informal)
[M-VLM2Vec: Structure-Preserving Matryoshka Representation Learning for Vision-Language Embeddings](https://aclanthology.org/2026.findings-emnlp.308/) (Do et al., Findings 2026)
ACL