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

• URL: http://arxiv.org/abs/2602.07026v1
• Date: Mon, 02 Feb 2026 13:59:39 GMT
• Title: Modality Gap-Driven Subspace Alignment Training Paradigm For Multimodal Large Language Models
• Authors: Xiaomin Yu, Yi Xin, Wenjie Zhang, Chonghan Liu, Hanzhen Zhao, Xiaoxing Hu, Xinlei Yu, Ziyue Qiao, Hao Tang, Xue Yang, Xiaobin Hu, Chengwei Qin, Hui Xiong, Yu Qiao, Shuicheng Yan,
• Abstract summary: A persistent geometric anomaly, the Modality Gap, remains.<n>Prior approaches to bridge this gap are largely limited by oversimplified isotropic assumptions.<n>We propose the Fixed-frame Modality Gap Theory, which decomposes the modality gap into stable biases and anisotropic residuals.<n>We then introduce ReAlign, a training-free modality alignment strategy.
• Score: 84.78794648147608
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• 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.

Related papers

• MultiModal Fine-tuning with Synthetic Captions [9.572235167281686]
  We propose a novel approach that transforms unimodal datasets into multimodal ones using Multimodal Large Language Models (MLLMs)<n>Our method employs carefully designed prompts incorporating class labels and domain context to produce high-quality captions for classification tasks.<n>Our work establishes a new paradigm for dataset enhancement that effectively bridges the gap between multimodal pre-training and fine-tuning.
  arXiv  Detail & Related papers  (2026-01-29T09:03:45Z)
• HyperET: Efficient Training in Hyperbolic Space for Multi-modal Large Language Models [50.31704374968706]
  Multi-modal large language models (MLLMs) have emerged as a transformative approach for aligning visual and textual understanding.<n>They typically require extremely high computational resources for training to achieve cross-modal alignment at multi-granularity levels.<n>We argue that a key source of this inefficiency lies in the vision encoders they widely equip with, e.g., CLIP and SAM, which lack the alignment with language at multi-granularity levels.
  arXiv  Detail & Related papers  (2025-10-23T08:16:44Z)
• MAGE: Multimodal Alignment and Generation Enhancement via Bridging Visual and Semantic Spaces [23.447713697204225]
  MAGE is a novel framework that bridges the semantic spaces of vision and text through an innovative alignment mechanism.<n>We employ a training strategy that combines cross-entropy and mean squared error, significantly enhancing the alignment effect.<n>Our proposed multimodal large model architecture, MAGE, achieved significantly better performance compared to similar works across various evaluation benchmarks.
  arXiv  Detail & Related papers  (2025-07-29T12:17:46Z)
• ViLa-MIL: Dual-scale Vision-Language Multiple Instance Learning for Whole Slide Image Classification [52.405499816861635]
  Multiple instance learning (MIL)-based framework has become the mainstream for processing the whole slide image (WSI)<n>We propose a dual-scale vision-language multiple instance learning (ViLa-MIL) framework for whole slide image classification.
  arXiv  Detail & Related papers  (2025-02-12T13:28:46Z)
• Gramian Multimodal Representation Learning and Alignment [5.793118803623239]
  We present the novel Gramian Representation Alignment Measure (GRAM)<n>GRAM learns and aligns $n$ modalities directly in the higher-dimensional space in which modality embeddings lie.<n>The novel GRAM-based contrastive loss function enhances the alignment of multimodal models in the higher-dimensional embedding space.
  arXiv  Detail & Related papers  (2024-12-16T16:41:51Z)
• FUSE-ing Language Models: Zero-Shot Adapter Discovery for Prompt Optimization Across Tokenizers [55.2480439325792]
  We propose FUSE, an approach to approximating an adapter layer that maps from one model's textual embedding space to another, even across different tokenizers. We show the efficacy of our approach via multi-objective optimization over vision-language and causal language models for image captioning and sentiment-based image captioning.
  arXiv  Detail & Related papers  (2024-08-09T02:16:37Z)
• ARNet: Self-Supervised FG-SBIR with Unified Sample Feature Alignment and Multi-Scale Token Recycling [11.129453244307369]
  FG-SBIR aims to minimize the distance between sketches and corresponding images in the embedding space.<n>We propose an effective approach to narrow the gap between the two domains.<n>It mainly facilitates unified mutual information sharing both intra- and inter-samples.
  arXiv  Detail & Related papers  (2024-06-17T13:49:12Z)
• Mitigating Modality Collapse in Multimodal VAEs via Impartial Optimization [7.4262579052708535]
  We argue that this effect is a consequence of conflicting gradients during multimodal VAE training. We show how to detect the sub-graphs in the computational graphs where gradients conflict. We empirically show that our framework significantly improves the reconstruction performance, conditional generation, and coherence of the latent space across modalities.
  arXiv  Detail & Related papers  (2022-06-09T13:29:25Z)
• Dense Contrastive Visual-Linguistic Pretraining [53.61233531733243]
  Several multimodal representation learning approaches have been proposed that jointly represent image and text. These approaches achieve superior performance by capturing high-level semantic information from large-scale multimodal pretraining. We propose unbiased Dense Contrastive Visual-Linguistic Pretraining to replace the region regression and classification with cross-modality region contrastive learning.
  arXiv  Detail & Related papers  (2021-09-24T07:20:13Z)
• Learning Deformable Image Registration from Optimization: Perspective, Modules, Bilevel Training and Beyond [62.730497582218284]
  We develop a new deep learning based framework to optimize a diffeomorphic model via multi-scale propagation. We conduct two groups of image registration experiments on 3D volume datasets including image-to-atlas registration on brain MRI data and image-to-image registration on liver CT data.
  arXiv  Detail & Related papers  (2020-04-30T03:23:45Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.