Multimodal Unlearning Across Vision, Language, Video, and Audio: Survey of Methods, Datasets, and Benchmarks
Nobin Sarwar, Shubhashis Roy Dipta, Zheyuan Liu, Vaidehi Patil
cs.LG, cs.AI, cs.CL, cs.CR, cs.MM
Submitted: 2026-07-08
Comments: Accepted to ACL Findings 2026
Project page: https://smsnobin77.github.io/Awesome-Multimodal-Unlearning
License: http://creativecommons.org/licenses/by/4.0/
The gist: 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
Terminology
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/
Sources
- Sealing The Backdoor: Unlearning Adversarial Text Triggers In Diffusion Models Using Knowledge Distillation
- MusicLM: Generating Music From Text
- Safety Mirage: How Spurious Correlations Undermine VLM Safety Fine-Tuning and Can Be Mitigated by Machine Unlearning
- MU-Bench: A Multitask Multimodal Benchmark for Machine Unlearning
- Speech Unlearning
- Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models
- SafeEraser: Enhancing Safety in Multimodal Large Language Models through Multimodal Machine Unlearning
- The Linear Geometry of Interpretable Tokens: Jailbreaking Attacks and Defenses for Unlearned Diffusion Models
- Training Data Attribution for Diffusion Models
- Human Motion Unlearning
- Rethinking Bottlenecks in Safety Fine-Tuning of Vision Language Models
- CLEAR: Character Unlearning in Textual and Visual Modalities
- A Comprehensive Survey of Machine Unlearning Techniques for Large Language Models
- Video Unlearning via Low-Rank Refusal Vector
- A Survey on Generative Model Unlearning: Fundamentals, Taxonomy, Evaluation, and Future Direction
- Probing Unlearned Diffusion Models: A Transferable Adversarial Attack Perspective
- Bridging Language and Items for Retrieval and Recommendation: Benchmarking LLMs as Semantic Encoders
- Backdoor Defense in Diffusion Models via Spatial Attention Unlearning
- Unconsciously Forget: Mitigating Memorization; Without Knowing What is being Memorized
- Enhancing User-Centric Privacy Protection: An Interactive Framework through Diffusion Models and Machine Unlearning
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks