mlsanityData to 14 Oct 2026

Meta-Unlearning on Diffusion Models: Preventing Relearning Unlearned Concepts

arXiv 2410.12777 ↗First posted 16 Oct 2024cs.CV cs.CL cs.CRcited by 40

On the map: Diffusion models → Robustness, privacy and adversarial aspects

also in 1 more
Abstract (arXiv)

With the rapid progress of diffusion-based content generation, significant efforts are being made to unlearn harmful or copyrighted concepts from pretrained diffusion models (DMs) to prevent potential model misuse. However, it is observed that even when DMs are properly unlearned before release, malicious finetuning can compromise this process, causing DMs to relearn the unlearned concepts. This occurs partly because certain benign concepts (e.g., "skin") retained in DMs are related to the unlearned ones (e.g., "nudity"), facilitating their relearning via finetuning. To address this, we propose meta-unlearning on DMs. Intuitively, a meta-unlearned DM should behave like an unlearned DM when used as is; moreover, if the meta-unlearned DM undergoes malicious finetuning on unlearned concepts, the related benign concepts retained within it will be triggered to self-destruct, hindering the relearning of unlearned concepts. Our meta-unlearning framework is compatible with most existing unlearning methods, requiring only the addition of an easy-to-implement meta objective. We validate our approach through empirical experiments on meta-unlearning concepts from Stable Diffusion models (SD-v1-4 and SDXL), supported by extensive ablation studies. Our code is available at https://github.com/sail-sg/Meta-Unlearning.

Key facts. Cited by 40 papers in the map (data through 14 Oct 2026). Placed in Robustness, privacy and adversarial aspects, one of 14 directions in Diffusion models. Builds on 67 of its 77 references that are in the map. First posted on arXiv 16 Oct 2024.

Builds on

67 of 77 references.

Closest ancestors · 19

Same direction, newest first

  1. 2024Reliable and Efficient Concept Erasure of Text-to-Image Diffusion Models17 Jul
  2. 2024R.A.C.E.: Robust Adversarial Concept Erasure for Secure Text-to-Image Diffusion Model25 May
  3. 2024Unlearning Concepts in Diffusion Model via Concept Domain Correction and Concept Preserving Gradient24 May
  4. 2024Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models24 May
  5. 2024Probing Unlearned Diffusion Models: A Transferable Adversarial Attack Perspective30 Apr
  6. 2024Concept Arithmetics for Circumventing Concept Inhibition in Diffusion Models21 Apr
  7. 2024Jailbreaking Prompt Attack: A Controllable Adversarial Attack against Diffusion Models2 Apr
  8. 2024MACE: Mass Concept Erasure in Diffusion Models10 Mar
  9. 2023Receler: Reliable Concept Erasing of Text-to-Image Diffusion Models via Lightweight Erasers29 Nov
  10. 2023To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images ... For Now18 Oct
  11. 2023Ring-A-Bell! How Reliable are Concept Removal Methods for Diffusion Models?16 Oct
  12. 2023Unified Concept Editing in Diffusion Models25 Aug
7 more
  1. 2023Towards Safe Self-Distillation of Internet-Scale Text-to-Image Diffusion Models12 Jul
  2. 2023Evaluating the Robustness of Text-to-image Diffusion Models against Real-world Attacks16 Jun
  3. 2023Forget-Me-Not: Learning to Forget in Text-to-Image Diffusion Models30 Mar
  4. 2023Ablating Concepts in Text-to-Image Diffusion Models23 Mar
  5. 2023Erasing Concepts from Diffusion Models13 Mar
  6. 2022Safe Latent Diffusion: Mitigating Inappropriate Degeneration in Diffusion Models9 Nov
  7. 2022Red-Teaming the Stable Diffusion Safety Filter3 Oct

Other references · 48

  1. 2020Denoising Diffusion Probabilistic Modelscited by 21,076
  2. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
  3. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,863
  4. 2014Microsoft COCO: Common Objects in Contextcited by 17,608
  5. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 9,122
  6. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,058
  7. 2017Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networkscited by 4,897
  8. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesiscited by 4,356
  9. 2017Prototypical Networks for Few-shot Learningcited by 3,015
  10. 2021CLIPScore: A Reference-free Evaluation Metric for Image Captioningcited by 2,424
  11. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 1,154
  12. 2018On First-Order Meta-Learning Algorithmscited by 1,018
36 more
  1. 2024TOFU: A Task of Fictitious Unlearning for LLMscited by 533
  2. 2019Meta-Learning with Implicit Gradientscited by 525
  3. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 509
  4. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearningcited by 504
  5. 2018Meta-Learning with Latent Embedding Optimizationcited by 476
  6. 2019Meta-Learning with Differentiable Convex Optimizationcited by 461
  7. 2021Multimodal datasets: misogyny, pornography, and malignant stereotypescited by 327
  8. 2023Large Language Model Unlearningcited by 302
  9. 2017Meta Networkscited by 269
  10. 2023In-Context Unlearning: Language Models as Few Shot Unlearnerscited by 224
  11. 2023Selective Amnesia: A Continual Learning Approach to Forgetting in Deep Generative Modelscited by 177
  12. 2023Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against Extraction Attackscited by 158
  13. 2024Towards Safer Large Language Models through Machine Unlearningcited by 139
  14. 2024Tamper-Resistant Safeguards for Open-Weight LLMscited by 138
  15. 2024An Adversarial Perspective on Machine Unlearning for AI Safetycited by 123
  16. 2024Machine Unlearning of Pre-trained Large Language Modelscited by 119
  17. 2024Large Language Model Unlearning via Embedding-Corrupted Promptscited by 116
  18. 2024SOUL: Unlocking the Power of Second-Order Optimization for LLM Unlearningcited by 104
  19. 2024Guardrail Baselines for Unlearning in LLMscited by 96
  20. 2024Reversing the Forget-Retain Objectives: An Efficient LLM Unlearning Framework from Logit Differencecited by 88
  21. 2023Circumventing Concept Erasure Methods For Text-to-Image Generative Modelscited by 85
  22. 2024GuardT2I: Defending Text-to-Image Models from Adversarial Promptscited by 62
  23. 2022Self-Destructing Models: Increasing the Costs of Harmful Dual Uses of Foundation Modelscited by 57
  24. 2024UnUnlearning: Unlearning is not sufficient for content regulation in advanced generative AIcited by 51
  25. 2024Direct Unlearning Optimization for Robust and Safe Text-to-Image Modelscited by 50
  26. 2024Soft Prompting for Unlearning in Large Language Modelscited by 46
  27. 2024Selective Forgetting: Advancing Machine Unlearning Techniques and Evaluation in Language Modelscited by 39
  28. 2024Offset Unlearning for Large Language Modelscited by 36
  29. 2024Fast Exact Unlearning for In-Context Learning Data for LLMscited by 34
  30. 2024Robust Concept Erasure Using Task Vectorscited by 27
  31. 2024Towards Robust Knowledge Unlearning: An Adversarial Framework for Assessing and Improving Unlearning Robustness in Large Language Modelscited by 27
  32. 2024To Forget or Not? Towards Practical Knowledge Unlearning for Large Language Modelscited by 26
  33. 2024RKLD: Reverse KL-Divergence-based Knowledge Distillation for Unlearning Personal Information in Large Language Modelscited by 21
  34. 2024Learning to Unlearn for Robust Machine Unlearningcited by 19
  35. 2024Second-Order Information Matters: Revisiting Machine Unlearning for Large Language Modelscited by 18
  36. 2024Learn while Unlearn: An Iterative Unlearning Framework for Generative Language Modelscited by 5

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 1; 2025: 14; 2026: 25.

40 papers.

Sort:

By area:

  1. 2024EraseAnything: Enabling Concept Erasure in Rectified Flow Transformerscited by 95
  2. 2025Set You Straight: Auto-Steering Denoising Trajectories to Sidestep Unwanted Conceptscited by 62
  3. 2025MMUnlearner: Reformulating Multimodal Machine Unlearning in the Era of Multimodal Large Language Modelscited by 30
  4. 2025ForgetMe: Evaluating Selective Forgetting in Generative Modelscited by 11
  5. 2025CTRAP: Embedding Collapse Trap to Safeguard Large Language Models from Harmful Fine-Tuningcited by 11
  6. 2025Erasing Concepts, Steering Generations: A Comprehensive Survey of Concept Suppressioncited by 10
  7. 2025Towards Resilient Safety-driven Unlearning for Diffusion Models against Downstream Fine-tuningcited by 9
  8. 2026Multimodal Unlearning Across Vision, Language, Video, and Audio: Survey of Methods, Datasets, and Benchmarkscited by 7
  9. 2025Locality-Aware Continual Unlearning for Diffusion Modelscited by 5
  10. 2026Differential Vector Erasure: Unified Training-Free Concept Erasure for Flow Matching Modelscited by 5
  11. 2025Unlearning's Blind Spots: Over-Unlearning and Prototypical Relearning Attackcited by 4
  12. 2025Reminiscence Attack on Residuals: Exploiting Approximate Machine Unlearning for Privacycited by 4
28 more
  1. 2026Z-Erase: Enabling Concept Erasure in Single-Stream Diffusion Transformerscited by 4
  2. 2025Perturb a Model, Not an Image: Towards Robust Privacy Protection via Anti-Personalized Diffusion Modelscited by 3
  3. 2026Models Designed to Forget: Machine Unlearning via Key Deletioncited by 3
  4. 2025Backdoor Defense in Diffusion Models via Spatial Attention Unlearningcited by 2
  5. 2025Not Every Time and Frequency Need to Be Forgotten in Diffusion Unlearningcited by 2
  6. 2026SafeRoPE: Risk-specific Head-wise Embedding Rotation for Safe Generation in Rectified Flow Transformerscited by 2
  7. 2026Unlearning in Diffusion Models: A Unified Framework with KL Divergence and Likelihood Constraintscited by 2
  8. 2025Comprehensive Evaluation and Analysis for NSFW Concept Erasure in Text-to-Image Diffusion Modelscited by 1
  9. 2026The Illusion of Forgetting: Attack Unlearned Diffusion via Initial Latent Variable Optimizationcited by 1
  10. 2026EraseAnything++: Enabling Concept Erasure in Rectified Flow Transformers Leveraging Multi-Object Optimizationcited by 1
  11. 2026Reference-Guided Machine Unlearningcited by 1
  12. 2026Evaluation without Generation: Non-Generative Assessment of Harmful Model Specialization with Applications to CSAMcited by 1
  13. 2026Your Unlearning Gives You Away: Identifying Erased Concepts in Diffusion Modelscited by 1
  14. 2025Comprehensive Assessment and Analysis for NSFW Content Erasure in Text-to-Image Diffusion Modelscited by 0
  15. 2026Why Instruction-Based Unlearning Fails in Diffusion Models?cited by 0
  16. 2026Projected Gradient Unlearning for Text-to-Image Diffusion Models: Defending Against Concept Revival Attackscited by 0
  17. 2026BARRIER: Bounded Activation Regions for Robust Information Erasurecited by 0
  18. 2026CoreUnlearn: Rethinking Concept Unlearning through Disentangled Component-Level Erasure in Text-guided Diffusion Modelscited by 0
  19. 2026Unified Safe In-context Image Generation in Multimodal Diffusion Transformers via Restricting Unsafe Information Flowscited by 0
  20. 2026IREU: Identity-Related Encoder-Only Unlearning for Customized Portrait Generationcited by 0
  21. 2026Position: Preventing AI-Generated CSAM Necessitates New Approaches to AI Safetycited by 0
  22. 2026TILDE: TILt-based Distributional Erasure for Concept Unlearningcited by 0
  23. 2026Stochastic Meta-Unlearning: Bridging Language Backbone and Multimodal Unlearningcited by 0
  24. 2026LU-500: A Logo Benchmark for Concept Unlearningcited by 0
  25. 2026IP Protection in the Era of Visual Generative AI: A Surveycited by 0
  26. 2026Mitigating Sequential Reappearance in Diffusion Data-Point Unlearningcited by 0
  27. 2026Continual Data Unlearning in Diffusion Models via Transition-based Regularizationcited by 0
  28. 2026You Can't Have It Both Ways: Concept Entanglement Limits Diffusion Model Unlearningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Unstable Unlearning: The Hidden Risk of Concept Resurgence in Diffusion Models
  2. 2024Holistic Unlearning Benchmark: A Multi-Faceted Evaluation for Text-to-Image Diffusion Model Unlearning
  3. 2026Co-occurring Associated REtained concepts in Diffusion Unlearning
  4. 2026Roots Beneath the Cut: Uncovering the Risk of Concept Revival in Pruning-Based Unlearning for Diffusion Models
  5. 2026eval-unlearn: Benchmarking unlearning in Text-to-Image Diffusion Models
  6. 2024Unlearning or Concealment? A Critical Analysis and Evaluation Metrics for Unlearning in Diffusion Models
  7. 2026Compensation-free Machine Unlearning in Text-to-Image Diffusion Models by Eliminating the Mutual Information
  8. 2026Erasure or Erosion? Evaluating Compositional Degradation in Unlearned Text-To-Image Diffusion Models