The Year in AI Papers
The year/Independent research

Paper 2602.07026

Modality Gap-Driven Subspace Alignment Training Paradigm For Multimodal Large Language Models

Xiaomin Yu, Yi Xin, Yuhui Zhang, Wenjie Zhang, Chonghan Liu, Hanzhen Zhao, Chen Liu, Xiaoxing Hu, Ziyue Qiao, Hao Tang, Xiaobin Hu, Chengwei Qin, et al.

Published
Feb 2026
Research lab
Independent
Citations
9
GitHub
76 stars

01 In brief

Summary

The paper addresses the Modality Gap in multimodal contrastive learning, where embeddings of different modalities for the same semantics occupy offset regions.

Prior methods rely on isotropic assumptions, which are flawed.

The authors propose the Fixed-frame Modality Gap Theory, decomposing the gap into stable biases (PMB, POB) and anisotropic residuals, plus a normalization-induced Phantom Drift.

Based on this, they introduce ReAlign, a training-free alignment strategy using three steps: Anchor Alignment (first-order mean), Trace Alignment (global variance), and Centroid Alignment (correcting spherical drift).

ReAlign uses statistics from unpaired data to map text representations into the image distribution.

They also propose ReVision, a two-stage MLLM training paradigm: Stage 1 uses ReAlign to convert unpaired text into pseudo-visual embeddings for pretraining, and Stage 2 uses real images for visual instruction tuning.

Experiments show ReAlign reduces the modality gap to 10^-4 scale (vs.

0.0023 for C3) and ReVision achieves an average score of 50.16, outperforming C3 (48.06) and paired-data training (48.91) at lower cost.

ReVision-2M surpasses the paired baseline at 74% of the cost, demonstrating that unpaired text can substitute for expensive image-text pairs in MLLM scaling.

02 From the paper

Abstract

Despite the success of multimodal contrastive learning in aligning visual and linguistic representations, a persistent geometric anomaly, the Modality Gap, remains: embeddings of distinct modalities expressing identical semantics occupy systematically offset regions. Prior approaches to bridge this gap are largely limited by oversimplified isotropic assumptions, hindering their application in large-scale scenarios. In this paper, we address these limitations by precisely characterizing the geometric shape of the modality gap and leveraging it for efficient model scaling. First, we propose the Fixed-frame Modality Gap Theory, which decomposes the modality gap within a frozen reference frame into stable biases and anisotropic residuals. Guided by this precise modeling, we introduce ReAlign, a training-free modality alignment strategy. Utilizing statistics from massive unpaired data, ReAlign aligns text representation into the image representation distribution via a three-step process comprising Anchor, Trace, and Centroid Alignment, thereby explicitly rectifying geometric misalignment. Building on ReAlign, we propose ReVision, a scalable training paradigm for Multimodal Large Language Models~(MLLMs). ReVision integrates ReAlign into the pretraining stage, enabling the model to learn the distribution of visual representations from unpaired text before visual instruction tuning, without the need for large-scale, high-quality image-text pairs. Our framework demonstrates that statistically aligned unpaired data can effectively substitute for expensive image-text pairs, offering a robust path for the efficient scaling of MLLMs.