Merge then Realign: Simple and Effective Modality-Incremental Continual Learning for Multimodal LLMs

• URL: http://arxiv.org/abs/2503.07663v2
• Date: Wed, 22 Oct 2025 08:23:29 GMT
• Title: Merge then Realign: Simple and Effective Modality-Incremental Continual Learning for Multimodal LLMs
• Authors: Dingkun Zhang, Shuhan Qi, Xinyu Xiao, Kehai Chen, Xuan Wang,
• Abstract summary: We propose an elegantly simple MCL paradigm called "MErge then ReAlign" (MERA) to address both forgetting and misalignment.<n>Experiments demonstrate the impressive performance of MERA, holding an average of 99.84% Backward Relative Gain when extending to four modalities.
• Score: 32.09688735548092
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Recent advances in Multimodal Large Language Models (MLLMs) have enhanced their versatility as they integrate a growing number of modalities. Considering the heavy cost of training MLLMs, it is efficient to reuse the existing ones and extend them to more modalities through Modality-incremental Continual Learning (MCL). The exploration of MCL is in its early stages. In this work, we dive into the causes of performance degradation in MCL. We uncover that it suffers not only from forgetting as in traditional continual learning, but also from misalignment between the modality-agnostic and modality-specific components. To this end, we propose an elegantly simple MCL paradigm called "MErge then ReAlign" (MERA) to address both forgetting and misalignment. MERA avoids introducing heavy model budgets or modifying model architectures, hence is easy to deploy and highly reusable in the MLLM community. Extensive experiments demonstrate the impressive performance of MERA, holding an average of 99.84\% Backward Relative Gain when extending to four modalities, achieving nearly lossless MCL performance. Our findings underscore the misalignment issue in MCL. More broadly, our work showcases how to adjust different components of MLLMs during continual learning.

Related papers

• Magic-MM-Embedding: Towards Visual-Token-Efficient Universal Multimodal Embedding with MLLMs [10.443777669301983]
  Multimodal Large Language Models (MLLMs) have shown immense promise in universal multimodal retrieval.<n>But their practical application is often hindered by the substantial computational cost incurred from processing a large number of tokens from visual inputs.<n>We propose Magic-MM-Embedding, a series of novel models that achieve both high efficiency and state-of-the-art performance in universal multimodal embedding.
  arXiv  Detail & Related papers  (2026-02-05T04:01:01Z)
• Mitigating Modality Imbalance in Multi-modal Learning via Multi-objective Optimization [57.00656508727821]
  Multi-modal learning (MML) aims to integrate information from multiple modalities, which is expected to lead to superior performance over single-modality learning.<n>Recent studies have shown that MML can underperform, even compared to single-modality approaches, due to imbalanced learning across modalities.<n>We propose a gradient-based algorithm to solve the modified MML problem.
  arXiv  Detail & Related papers  (2025-11-10T04:16:01Z)
• NaViL: Rethinking Scaling Properties of Native Multimodal Large Language Models under Data Constraints [100.02131897927484]
  This paper focuses on the native training of Multimodal Large Language Models (MLLMs) in an end-to-end manner.<n>We propose a native MLLM called NaViL, combined with a simple and cost-effective recipe.<n> Experimental results on 14 multimodal benchmarks confirm the competitive performance of NaViL against existing MLLMs.
  arXiv  Detail & Related papers  (2025-10-09T17:59:37Z)
• PUMA: Layer-Pruned Language Model for Efficient Unified Multimodal Retrieval with Modality-Adaptive Learning [54.73049408950049]
  We propose a Layer-Pruned Language Model for Efficient Unified Multimodal Retrieval with Modality-Adaptive Learning.<n>Our approach improves unified multimodal retrieval from both structural and learning perspectives.
  arXiv  Detail & Related papers  (2025-07-10T16:47:25Z)
• Meta-Adaptive Prompt Distillation for Few-Shot Visual Question Answering [53.82094608038132]
  Large Multimodal Models (LMMs) often rely on in-context learning (ICL) to perform new tasks with minimal supervision.<n>ICL performance, especially in smaller LMMs, is inconsistent and does not always improve monotonically with increasing examples.<n>We propose a meta-learning approach that provides an alternative for inducing few-shot capabilities in LMMs.
  arXiv  Detail & Related papers  (2025-06-07T19:37:22Z)
• MLLM-CL: Continual Learning for Multimodal Large Language Models [62.90736445575181]
  We introduce MLLM-CL, a novel benchmark encompassing domain and ability continual learning.<n>Our approach can integrate domain-specific knowledge and functional abilities with minimal forgetting, significantly outperforming existing methods.
  arXiv  Detail & Related papers  (2025-06-05T17:58:13Z)
• Reasoning-Aligned Perception Decoupling for Scalable Multi-modal Reasoning [95.44766931218896]
  Multi-modal large language models (MLLMs) still lag behind text-based reasoning.<n>We introduce Perception-Reasoning Decoupling, which modularizes the MLLM's reasoning component and makes it easily replaceable.<n>We propose a novel reinforcement learning algorithm called Visual Perception Optimization (VPO) to align the MLLM's perceptual output with the final reasoning task.
  arXiv  Detail & Related papers  (2025-06-05T02:28:07Z)
• Improve Multi-Modal Embedding Learning via Explicit Hard Negative Gradient Amplifying [7.9925771591348065]
  Core contrastive learning paradigm remains largely unchanged from CLIP-style models to MLLMs.<n>In this work, we conduct a detailed analysis of the gradients of the info-NCE loss with respect to the query, positive, and negative samples.<n>We propose to explicitly amplify the gradients associated with hard negative samples, thereby encouraging the model to learn more discriminative embeddings.
  arXiv  Detail & Related papers  (2025-05-28T11:18:19Z)
• Multi-Modality Expansion and Retention for LLMs through Parameter Merging and Decoupling [33.90190836855285]
  MMER integrates existing MLLMs for effective multimodal expansion while retaining their original performance.<n>MMER can also mitigate catastrophic forgetting by applying a similar process to MLLMs fine-tuned on new tasks.
  arXiv  Detail & Related papers  (2025-05-21T12:40:07Z)
• Orchestrate Multimodal Data with Batch Post-Balancing to Accelerate Multimodal Large Language Model Training [12.911726316306755]
  We introduce OrchMLLM, a framework designed to mitigate the inefficiencies in MLLM training caused by Modality Composition Incoherence.<n> Batch Post-Balancing Dispatcher and MLLM Global Orchestrator are used to eliminate mini-batch imbalances in sequential data.<n>OrchMLLM achieves a Model FLOPs Utilization (MFU) of $41.6%$ when training an 84B MLLM with three modalities on $2560$ H100 GPU, outperforming Megatron-LM by up to $3.1times$ in throughput.
  arXiv  Detail & Related papers  (2025-03-31T08:24:23Z)
• PAL: Prompting Analytic Learning with Missing Modality for Multi-Modal Class-Incremental Learning [42.00851701431368]
  Multi-modal class-incremental learning (MMCIL) seeks to leverage multi-modal data, such as audio-visual and image-text pairs.<n>A critical challenge remains: the issue of missing modalities during incremental learning phases.<n>We propose PAL, a novel exemplar-free framework tailored to MMCIL under missing-modality scenarios.
  arXiv  Detail & Related papers  (2025-01-16T08:04:04Z)
• Modality-Inconsistent Continual Learning of Multimodal Large Language Models [37.15220266767881]
  We introduce Modality-Inconsistent Continual Learning (MICL), a new continual learning scenario for Multimodal Large Language Models (MLLMs)<n>Unlike existing vision-only or modality-incremental settings, MICL combines modality and task type shifts, both of which drive catastrophic forgetting.<n>We propose MoInCL, which employs a Pseudo Targets Generation Module to mitigate forgetting caused by task type shifts in previously seen modalities.
  arXiv  Detail & Related papers  (2024-12-17T16:13:56Z)
• Can MLLMs Guide Weakly-Supervised Temporal Action Localization Tasks? [6.7065734065794835]
  We introduce a novel learning paradigm termed MLLM4WTAL. It harnesses the potential of MLLM to offer temporal action key semantics and complete semantic priors. It achieves this by integrating two distinct modules: Key Semantic Matching (KSM) and Complete Semantic Reconstruction (CSR)
  arXiv  Detail & Related papers  (2024-11-13T09:37:24Z)
• Classifier-guided Gradient Modulation for Enhanced Multimodal Learning [50.7008456698935]
  Gradient-Guided Modulation (CGGM) is a novel method to balance multimodal learning with gradients. We conduct extensive experiments on four multimodal datasets: UPMC-Food 101, CMU-MOSI, IEMOCAP and BraTS. CGGM outperforms all the baselines and other state-of-the-art methods consistently.
  arXiv  Detail & Related papers  (2024-11-03T02:38:43Z)
• LLMs Can Evolve Continually on Modality for X-Modal Reasoning [62.2874638875554]
  Existing methods rely heavily on modal-specific pretraining and joint-modal tuning, leading to significant computational burdens when expanding to new modalities. We propose PathWeave, a flexible and scalable framework with modal-Path sWitching and ExpAnsion abilities. PathWeave performs comparably to state-of-the-art MLLMs while concurrently reducing parameter training burdens by 98.73%.
  arXiv  Detail & Related papers  (2024-10-26T13:19:57Z)
• LLaVA-KD: A Framework of Distilling Multimodal Large Language Models [72.68665884790002]
  We propose a novel framework to transfer knowledge from l-MLLMs to s-MLLMs.<n>We introduce Multimodal Distillation (MDist) to transfer teacher model's robust representations across both visual and linguistic modalities.<n>We also propose a three-stage training scheme to fully exploit the potential of the proposed distillation strategy.
  arXiv  Detail & Related papers  (2024-10-21T17:41:28Z)
• Mono-InternVL: Pushing the Boundaries of Monolithic Multimodal Large Language Models with Endogenous Visual Pre-training [48.455597568212944]
  We present Mono-InternVL, a novel monolithic MLLM that seamlessly integrates a set of visual experts via a multimodal mixture-of-experts structure. In particular, EViP is designed as a progressive learning process for visual experts, which aims to fully exploit the visual knowledge from noisy data to high-quality data.
  arXiv  Detail & Related papers  (2024-10-10T17:59:22Z)
• MoExtend: Tuning New Experts for Modality and Task Extension [61.29100693866109]
  MoExtend is an effective framework designed to streamline the modality adaptation and extension of Mixture-of-Experts (MoE) models. MoExtend seamlessly integrates new experts into pre-trained MoE models, endowing them with novel knowledge without the need to tune pretrained models.
  arXiv  Detail & Related papers  (2024-08-07T02:28:37Z)
• Model Composition for Multimodal Large Language Models [71.5729418523411]
  We propose a new paradigm through the model composition of existing MLLMs to create a new model that retains the modal understanding capabilities of each original model. Our basic implementation, NaiveMC, demonstrates the effectiveness of this paradigm by reusing modality encoders and merging LLM parameters.
  arXiv  Detail & Related papers  (2024-02-20T06:38:10Z)
• Improving Discriminative Multi-Modal Learning with Large-Scale Pre-Trained Models [51.5543321122664]
  This paper investigates how to better leverage large-scale pre-trained uni-modal models to enhance discriminative multi-modal learning. We introduce Multi-Modal Low-Rank Adaptation learning (MMLoRA)
  arXiv  Detail & Related papers  (2023-10-08T15:01:54Z)
• Beyond Task Performance: Evaluating and Reducing the Flaws of Large Multimodal Models with In-Context Learning [105.77733287326308]
  We evaluate 10 recent open-source LMMs from 3B up to 80B parameter scale, on 5 different axes; hallucinations, abstention, compositionality, explainability and instruction following. We explore the training-free in-context learning (ICL) as a solution, and study how it affects these limitations. Based on our ICL study, (3) we push ICL further and propose new multimodal ICL variants such as; Multitask-ICL, Chain-of-Hindsight-ICL, and Self-Correcting-ICL.
  arXiv  Detail & Related papers  (2023-10-01T12:02:59Z)
• PMR: Prototypical Modal Rebalance for Multimodal Learning [11.5547414386921]
  We propose Prototypical Modality Rebalance (PMR) to perform stimulation on the particular slow-learning modality without interference from other modalities. Our method only relies on the representations of each modality and without restrictions from model structures and fusion methods.
  arXiv  Detail & Related papers  (2022-11-14T03:36:05Z)

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.