Multimodal Unlearning Across Vision, Language, Video, and Audio: Survey of Methods, Datasets, and Benchmarks

2026-07-08

AI Summary

This work addresses the challenge that multimodal foundation models often inadvertently encode sensitive or harmful cross-modal associations during training, and such knowledge is difficult to precisely remove without full retraining. The paper proposes the first systematic taxonomy for multimodal unlearning, comprehensively reviewing and comparing unlearning approaches across vision, language, audio, and video modalities—spanning techniques based on fine-tuning, regularization, representation disentanglement, and generative adversarial methods. The study establishes a unified evaluation benchmark and an open-source repository, explicitly characterizing the trade-offs among deletion strength, preservation of model utility, and computational efficiency. This contribution provides both theoretical grounding and practical guidance for advancing research and applications in multimodal unlearning.

Abstract

With the growing adoption of VLMs, DMs, LLMs, and AFMs, these multimodal foundation models can inadvertently encode sensitive, copyrighted, biased, or unsafe cross-modal associations that originate from their training data. Retraining after deletion requests or policy updates is often impractical, and targeted forgetting remains difficult because knowledge is distributed across shared representations. Multimodal unlearning addresses this challenge by enabling selective removal across modalities while retaining overall utility. This survey offers a unified, system-oriented view of multimodal unlearning across vision, language, audio, and video, grounded in recent advances, emerging applications, and open problems. Our taxonomy enables systematic comparison across model architectures and modalities, clarifying trade-offs among deletion strength, retention, efficiency, reversibility, and robustness. This survey highlights open problems and practical considerations to support future research and deployment of multimodal unlearning. We release a curated repository: https://smsnobin77.github.io/Awesome-Multimodal-Unlearning/

Problem

• multimodal unlearning
• foundation models
• data deletion
• cross-modal associations
• selective forgetting

Innovation

• multimodal unlearning
• foundation models
• cross-modal forgetting
• systematic taxonomy
• data deletion