mlsanityData to 14 Oct 2026

Unlearning Concepts in Diffusion Model via Concept Domain Correction and Concept Preserving Gradient

arXiv 2405.15304 ↗First posted 24 May 2024cs.LG cs.CVcited by 69

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

also in 1 more
Abstract (arXiv)

Text-to-image diffusion models have achieved remarkable success in generating photorealistic images. However, the inclusion of sensitive information during pre-training poses significant risks. Machine Unlearning (MU) offers a promising solution to eliminate sensitive concepts from these models. Despite its potential, existing MU methods face two main challenges: 1) limited generalization, where concept erasure is effective only within the unlearned set, failing to prevent sensitive concept generation from out-of-set prompts; and 2) utility degradation, where removing target concepts significantly impacts the model's overall performance. To address these issues, we propose a novel concept domain correction framework named DoCo (Domain Correction). By aligning the output domains of sensitive and anchor concepts through adversarial training, our approach ensures comprehensive unlearning of target concepts. Additionally, we introduce a concept-preserving gradient surgery technique that mitigates conflicting gradient components, thereby preserving the model's utility while unlearning specific concepts. Extensive experiments across various instances, styles, and offensive concepts demonstrate the effectiveness of our method in unlearning targeted concepts with minimal impact on related concepts, outperforming previous approaches even for out-of-distribution prompts.

Key facts. Cited by 69 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 21 of its 54 references that are in the map. First posted on arXiv 24 May 2024.

Builds on

21 of 54 references.

Closest ancestors · 5

Same direction, newest first

  1. 2023One-Dimensional Adapter to Rule Them All: Concepts, Diffusion Models and Erasing Applications26 Dec
  2. 2023To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images ... For Now18 Oct
  3. 2023Unified Concept Editing in Diffusion Models25 Aug
  4. 2023Extracting Training Data from Diffusion Models30 Jan
  5. 2022Safe Latent Diffusion: Mitigating Inappropriate Degeneration in Diffusion Models9 Nov

Other references · 16

  1. 2020Denoising Diffusion Probabilistic Modelscited by 21,076
  2. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,863
  3. 2020Denoising Diffusion Implicit Modelscited by 9,781
  4. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,355
  5. 2022Generative Adversarial Networkscited by 5,627
  6. 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,152
  7. 2021GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Modelscited by 3,614
  8. 2015Domain-Adversarial Training of Neural Networkscited by 3,044
  9. 2021CLIPScore: A Reference-free Evaluation Metric for Image Captioningcited by 2,424
  10. 2016Membership Inference Attacks against Machine Learning Modelscited by 2,281
  11. 2020Gradient Surgery for Multi-Task Learningcited by 1,256
  12. 2022Prompt-aligned Gradient for Prompt Tuningcited by 341
4 more
  1. 2022Membership Inference Attacks Against Text-to-image Generation Modelscited by 69
  2. 2022Generated Faces in the Wild: Quantitative Comparison of Stable Diffusion, Midjourney and DALL-E 2cited by 55
  3. 2024Endowing Pre-trained Graph Models with Provable Fairnesscited by 10
  4. 2023Self-Supervised 3D Scene Flow Estimation and Motion Prediction using Local Rigidity Priorcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 9; 2025: 32; 2026: 28.

69 papers.

Sort:

By area:

  1. 2024SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generationcited by 133
  2. 2025Safety at Scale: A Comprehensive Survey of Large Model and Agent Safetycited by 67
  3. 2024Meta-Unlearning on Diffusion Models: Preventing Relearning Unlearned Conceptscited by 40
  4. 2025SPEED: Scalable, Precise, and Efficient Concept Erasure for Diffusion Modelscited by 36
  5. 2024Adversarial Attacks and Defenses on Text-to-Image Diffusion Models: A Surveycited by 35
  6. 2024Attacks and Defenses for Generative Diffusion Models: A Comprehensive Surveycited by 32
  7. 2024Jailbreak Attacks and Defenses against Multimodal Generative Models: A Surveycited by 32
  8. 2025CURE: Concept Unlearning via Orthogonal Representation Editing in Diffusion Modelscited by 27
  9. 2025Minimalist Concept Erasure in Generative Modelscited by 13
  10. 2024On the Limitations and Prospects of Machine Unlearning for Generative AIcited by 12
  11. 2025Erasing Concepts, Steering Generations: A Comprehensive Survey of Concept Suppressioncited by 10
  12. 2025Image Can Bring Your Memory Back: A Novel Multi-Modal Guided Attack against Image Generation Model Unlearningcited by 10
38 more
  1. 2025Continual Unlearning for Text-to-Image Diffusion Models: A Regularization Perspectivecited by 10
  2. 2025CE-SDWV: Effective and Efficient Concept Erasure for Text-to-Image Diffusion Models via a Semantic-Driven Word Vocabularycited by 8
  3. 2024Model Integrity when Unlearning with T2I Diffusion Modelscited by 7
  4. 2024AEIOU: A Unified Defense Framework against NSFW Prompts in Text-to-Image Modelscited by 7
  5. 2025Position: Bridge the Gaps between Machine Unlearning and AI Regulationcited by 7
  6. 2025TarPro: Targeted Protection against Malicious Image Editingcited by 7
  7. 2026Look Carefully: Adaptive Visual Reinforcements in Multimodal Large Language Models for Hallucination Mitigationcited by 7
  8. 2026OrthoEraser: Coupled-Neuron Orthogonal Projection for Concept Erasurecited by 7
  9. 2026Multimodal Unlearning Across Vision, Language, Video, and Audio: Survey of Methods, Datasets, and Benchmarkscited by 7
  10. 2025Plug-and-Play Interpretable Responsible Text-to-Image Generation via Dual-Space Multi-facet Concept Controlcited by 5
  11. 2025Efficient Utility-Preserving Machine Unlearning with Implicit Gradient Surgerycited by 5
  12. 2025PurifyGen: A Risk-Discrimination and Semantic-Purification Model for Safe Text-to-Image Generationcited by 5
  13. 2025Rethinking Byzantine Robustness in Federated Recommendation from Sparse Aggregation Perspectivecited by 4
  14. 2025Concept Lancet: Image Editing with Compositional Representation Transplantcited by 4
  15. 2025Now You See It, Now You Don't - Instant Concept Erasure for Safe Text-to-Image and Video Generationcited by 4
  16. 2025Wukong Framework for Not Safe For Work Detection in Text-to-Image systemscited by 3
  17. 2025Collapse of Irrelevant Representations (CIR) Ensures Robust and Non-Disruptive LLM Unlearningcited by 3
  18. 2025Hawk: Leveraging Spatial Context for Faster Autoregressive Text-to-Image Generationcited by 3
  19. 2026SafeRedir: Prompt Embedding Redirection for Robust Unlearning in Image Generation Modelscited by 3
  20. 2026Conditioned Activation Transport for T2I Safety Steeringcited by 3
  21. 2025Backdoor Defense in Diffusion Models via Spatial Attention Unlearningcited by 2
  22. 2025Sealing The Backdoor: Unlearning Adversarial Text Triggers In Diffusion Models Using Knowledge Distillationcited by 2
  23. 2025A Unified Framework for Diffusion Model Unlearning with f-Divergencecited by 2
  24. 2025From Unlearning to UNBRANDING: A Benchmark for Trademark-Safe Text-to-Image Generationcited by 2
  25. 2026SafeRoPE: Risk-specific Head-wise Embedding Rotation for Safe Generation in Rectified Flow Transformerscited by 2
  26. 2025FALCON: Fine-grained Activation Manipulation by Contrastive Orthogonal Unalignment for Large Language Modelcited by 1
  27. 2025Do Concept Replacement Techniques Really Erase Unacceptable Concepts?cited by 1
  28. 2025GIFT: Gradient-aware Immunization of diffusion models against malicious Fine-Tuning with safe concepts retentioncited by 1
  29. 2025Beyond Fixed Anchors: Precisely Erasing Concepts with Sibling Exclusive Counterpartscited by 1
  30. 2025Rethinking Robust Adversarial Concept Erasure in Diffusion Modelscited by 1
  31. 2026Easy to Learn, Yet Hard to Forget: Towards Robust Unlearning Under Biascited by 1
  32. 2026IDDM: Identity-Decoupled Personalized Diffusion Models with a Tunable Privacy-Utility Trade-offcited by 1
  33. 2026Beyond Text Prompts: Precise Concept Erasure through Text-Image Collaborationcited by 1
  34. 2026Empty SPACE: Cross-Attention Sparsity for Concept Erasure in Diffusion Modelscited by 1
  35. 2024FameBias: Embedding Manipulation Bias Attack in Text-to-Image Modelscited by 0
  36. 2025Bridging Domain Gaps between Pretrained Multimodal Models and Recommendationscited by 0
  37. 2025SafeCtrl: Region-Based Safety Control for Text-to-Image Diffusion via Detect-Then-Suppresscited by 0
  38. 2025Metaphor-based Jailbreak Attacks on Text-to-Image Modelscited by 0

50 most cited of 69 All 69 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026Forget Many, Forget Right: Scalable and Precise Concept Unlearning in Diffusion Models
  2. 2024ConceptPrune: Concept Editing in Diffusion Models via Skilled Neuron Pruning
  3. 2024Unlearning or Concealment? A Critical Analysis and Evaluation Metrics for Unlearning in Diffusion Models
  4. 2024Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models
  5. 2024Unstable Unlearning: The Hidden Risk of Concept Resurgence in Diffusion Models
  6. 2024Erasing Concepts from Text-to-Image Diffusion Models with Few-shot Unlearning
  7. 2026Weeding Out Bad Seeds: Initial-Noise-Robust Unlearning for Text-to-Image Diffusion Models
  8. 2026eval-unlearn: Benchmarking unlearning in Text-to-Image Diffusion Models