Anisotropic Modality Align

Hugging Face Daily Papers Papers

Summary

This paper proposes AnisoAlign, a framework that addresses the modality gap in multimodal models by applying anisotropic geometric correction to enable effective unpaired modality alignment.

Training multimodal large language models has long been limited by the scarcity of high-quality paired multimodal data. Recent studies show that the shared representation space of pretrained multimodal contrastive models can serve as a bridge, enabling models to perform multimodal training with unimodal data. However, the key premise of this paradigm remains insufficiently understood: can representations from different modalities be reliably interchanged? The core obstacle lies in the persistent Modality Gap in the shared space. In this work, we revisit the geometric nature of the modality gap. We find that modality representations already share compatible dominant semantic geometry. What truly hinders modality interchangeability is not a simple global shift, but an anisotropic residual structure concentrated along a small number of dominant directions. Based on this finding, we further propose the principle of anisotropic modality gap alignment: effective modality alignment should align with the target-modality distribution while preserving the semantic structure of the source modality. Guided by this principle, we propose an anisotropic geometric correction framework, AnisoAlign, for unpaired modality alignment. This framework leverages the internal geometric prior of the target modality and performs bounded correction on source-modality representations, thereby constructing substitute representations in the target modality. Experiments confirm its benefits in both geometric diagnostics and text-only MLLM training. Overall, this work recasts the modality gap from an empirical observation into a correctable, structured geometric phenomenon and provides a new representation alignment perspective for training multimodal models with unimodal data.
Original Article
View Cached Full Text

Cached at: 05/11/26, 07:20 AM

Paper page - Anisotropic Modality Align

Source: https://huggingface.co/papers/2605.07825 Published on May 8

·

Submitted byhttps://huggingface.co/Yu2020

Yu_xmon May 11

Authors:

,

,

,

,

,

,

,

,

,

Abstract

Research addresses the modality gap in multimodal models by proposing an anisotropic geometric correction framework that enables effective unpaired modality alignment through structured representation transformation.

Trainingmultimodal large language modelshas long been limited by the scarcity of high-quality paired multimodal data. Recent studies show that theshared representation spaceof pretrainedmultimodal contrastive modelscan serve as a bridge, enabling models to perform multimodal training with unimodal data. However, the key premise of this paradigm remains insufficiently understood: can representations from different modalities be reliably interchanged? The core obstacle lies in the persistentModality Gapin the shared space. In this work, we revisit the geometric nature of themodality gap. We find that modality representations already share compatible dominantsemantic geometry. What truly hinders modality interchangeability is not a simple global shift, but ananisotropic residual structureconcentrated along a small number of dominant directions. Based on this finding, we further propose the principle of anisotropicmodality gapalignment: effectivemodality alignmentshould align with the target-modality distribution while preserving the semantic structure of the source modality. Guided by this principle, we propose an anisotropicgeometric correction framework, AnisoAlign, for unpairedmodality alignment. This framework leverages theinternal geometric priorof the target modality and performs bounded correction on source-modality representations, thereby constructing substitute representations in the target modality. Experiments confirm its benefits in both geometric diagnostics and text-only MLLM training. Overall, this work recasts themodality gapfrom an empirical observation into a correctable, structured geometric phenomenon and provides a new representation alignment perspective for training multimodal models with unimodal data.

View arXiv pageView PDFGitHub62Add to collection

Get this paper in your agent:

hf papers read 2605\.07825

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2605.07825 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.07825 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.07825 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

PolyAlign: Conditional Human-Distribution Alignment

arXiv cs.CL

PolyAlign is a distribution-aware alignment framework that aligns language models to context-specific human response distributions rather than a single global style, improving naturalness and faithfulness across bilingual settings.