Towards Uniformity and Alignment for Multimodal Representation Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yin, Wenzhe, Zhou, Pan, Xiao, Zehao, Liu, Jie, Yu, Shujian, Sonke, Jan-Jakob, Gavves, Efstratios
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912894085496832
author Yin, Wenzhe
Zhou, Pan
Xiao, Zehao
Liu, Jie
Yu, Shujian
Sonke, Jan-Jakob
Gavves, Efstratios
author_facet Yin, Wenzhe
Zhou, Pan
Xiao, Zehao
Liu, Jie
Yu, Shujian
Sonke, Jan-Jakob
Gavves, Efstratios
contents Multimodal representation learning aims to construct a shared embedding space in which heterogeneous modalities are semantically aligned. Despite strong empirical results, InfoNCE-based objectives introduce inherent conflicts that yield distribution gaps across modalities. In this work, we identify two conflicts in the multimodal regime, both exacerbated as the number of modalities increases: (i) an alignment-uniformity conflict, whereby the repulsion of uniformity undermines pairwise alignment, and (ii) an intra-alignment conflict, where aligning multiple modalities induces competing alignment directions. To address these issues, we propose a principled decoupling of alignment and uniformity for multimodal representations, providing a conflict-free recipe for multimodal learning that simultaneously supports discriminative and generative use cases without task-specific modules. We then provide a theoretical guarantee that our method acts as an efficient proxy for a global Hölder divergence over multiple modality distributions, and thus reduces the distribution gap among modalities. Extensive experiments on retrieval and UnCLIP-style generation demonstrate consistent gains.
format Preprint
id arxiv_https___arxiv_org_abs_2602_09507
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Towards Uniformity and Alignment for Multimodal Representation Learning
Yin, Wenzhe
Zhou, Pan
Xiao, Zehao
Liu, Jie
Yu, Shujian
Sonke, Jan-Jakob
Gavves, Efstratios
Machine Learning
Multimodal representation learning aims to construct a shared embedding space in which heterogeneous modalities are semantically aligned. Despite strong empirical results, InfoNCE-based objectives introduce inherent conflicts that yield distribution gaps across modalities. In this work, we identify two conflicts in the multimodal regime, both exacerbated as the number of modalities increases: (i) an alignment-uniformity conflict, whereby the repulsion of uniformity undermines pairwise alignment, and (ii) an intra-alignment conflict, where aligning multiple modalities induces competing alignment directions. To address these issues, we propose a principled decoupling of alignment and uniformity for multimodal representations, providing a conflict-free recipe for multimodal learning that simultaneously supports discriminative and generative use cases without task-specific modules. We then provide a theoretical guarantee that our method acts as an efficient proxy for a global Hölder divergence over multiple modality distributions, and thus reduces the distribution gap among modalities. Extensive experiments on retrieval and UnCLIP-style generation demonstrate consistent gains.
title Towards Uniformity and Alignment for Multimodal Representation Learning
topic Machine Learning
url https://arxiv.org/abs/2602.09507