mlsanityData to 7 Oct 2026

Data Attribution for Text-to-Image Models by Unlearning Synthesized Images

arXiv 2406.09408 ↗First posted 13 Jun 2024cs.CV cs.LGcited by 22

On the map: LLM safety and security → Vision-language and text-to-image safety

Abstract (arXiv)

The goal of data attribution for text-to-image models is to identify the training images that most influence the generation of a new image. Influence is defined such that, for a given output, if a model is retrained from scratch without the most influential images, the model would fail to reproduce the same output. Unfortunately, directly searching for these influential images is computationally infeasible, since it would require repeatedly retraining models from scratch. In our work, we propose an efficient data attribution method by simulating unlearning the synthesized image. We achieve this by increasing the training loss on the output image, without catastrophic forgetting of other, unrelated concepts. We then identify training images with significant loss deviations after the unlearning process and label these as influential. We evaluate our method with a computationally intensive but "gold-standard" retraining from scratch and demonstrate our method's advantages over previous methods.

Key facts. Cited by 22 papers in the map (data through 7 Oct 2026). Placed in Vision-language and text-to-image safety, one of 12 directions in LLM safety and security. Builds on 69 of its 75 references that are in the map. First posted on arXiv 13 Jun 2024.

Builds on

69 of 75 references.

Closest ancestors · 1

Same direction, newest first

  1. 2023Forget-Me-Not: Learning to Forget in Text-to-Image Diffusion Models30 Mar

Other references · 68

  1. 2017Attention Is All You Needcited by 59,932
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  3. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  4. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
  5. 2014Microsoft COCO: Common Objects in Contextcited by 17,498
  6. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,764
  7. 2021Emerging Properties in Self-Supervised Vision Transformerscited by 6,893
  8. 2022Hierarchical Text-Conditional Image Generation with CLIP Latentscited by 6,539
  9. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,322
  10. 2016Overcoming catastrophic forgetting in neural networkscited by 4,718
  11. 2014Neural Machine Translation by Jointly Learning to Align and Translatecited by 4,530
  12. 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,131
38 more
  1. 2021GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Modelscited by 3,595
  2. 2022DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generationcited by 3,286
  3. 2023Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detectioncited by 3,274
  4. 2017Understanding Black-box Predictions via Influence Functionscited by 1,714
  5. 2022Scaling Autoregressive Models for Content-Rich Text-to-Image Generationcited by 1,287
  6. 2022Multi-Concept Customization of Text-to-Image Diffusioncited by 1,099
  7. 2019Machine Unlearningcited by 1,020
  8. 2015Optimizing Neural Networks with Kronecker-factored Approximate Curvaturecited by 776
  9. 2023Extracting Training Data from Diffusion Modelscited by 723
  10. 2023Muse: Text-To-Image Generation via Masked Generative Transformerscited by 640
  11. 2019Data Shapley: Equitable Valuation of Data for Machine Learningcited by 617
  12. 2019Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Networkscited by 599
  13. 2020Estimating Training Data Influence by Tracing Gradient Descentcited by 565
  14. 2023Erasing Concepts from Diffusion Modelscited by 530
  15. 2019Certified Data Removal from Machine Learning Modelscited by 512
  16. 2020What Neural Networks Memorize and Why: Discovering the Long Tail via Influence Estimationcited by 492
  17. 2023Scaling up GANs for Text-to-Image Synthesiscited by 492
  18. 2022Make-A-Scene: Scene-Based Text-to-Image Generation with Human Priorscited by 487
  19. 2014New insights and perspectives on the natural gradient methodcited by 435
  20. 2022Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Modelscited by 418
  21. 2023Unified Concept Editing in Diffusion Modelscited by 377
  22. 2021Remember What You Want to Forget: Algorithms for Machine Unlearningcited by 360
  23. 2023Ablating Concepts in Text-to-Image Diffusion Modelscited by 312
  24. 2023Studying Large Language Model Generalization with Influence Functionscited by 311
  25. 2020Amnesiac Machine Learningcited by 305
  26. 2019Towards Efficient Data Valuation Based on the Shapley Valuecited by 279
  27. 2023TRAK: Attributing Model Behavior at Scalecited by 273
  28. 2022Dual Diffusion Implicit Bridges for Image-to-Image Translationcited by 212
  29. 2021Fast Yet Effective Machine Unlearningcited by 209
  30. 2023Key-Locked Rank One Editing for Text-to-Image Personalizationcited by 194
  31. 2022Datamodels: Predicting Predictions from Training Datacited by 186
  32. 2023Fast Machine Unlearning Without Retraining Through Selective Synaptic Dampeningcited by 177
  33. 2020Mixed-Privacy Forgetting in Deep Networkscited by 158
  34. 2013Revisiting Natural Gradient for Deep Networkscited by 157
  35. 2018Fast Approximate Natural Gradient Descent in a Kronecker-factored Eigenbasiscited by 143
  36. 2021Scaling Up Influence Functionscited by 130
  37. 2020Few-shot Image Generation with Elastic Weight Consolidationcited by 122
  38. 2023DataInf: Efficiently Estimating Data Influence in LoRA-tuned LLMs and Diffusion Modelscited by 121

50 most cited of 68 All 68 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 3; 2025: 9; 2026: 10.

22 papers.

Sort:

By area:

  1. 2025EraseFlow: Learning Concept Erasure Policies via GFlowNet-Driven Alignmentcited by 8
  2. 2024A LoRA is Worth a Thousand Picturescited by 7
  3. 2025Fast Data Attribution for Text-to-Image Modelscited by 7
  4. 2025Position: Bridge the Gaps between Machine Unlearning and AI Regulationcited by 5
  5. 2024Final-Model-Only Data Attribution with a Unifying View of Gradient-Based Methodscited by 4
  6. 2024IntroStyle: Training-Free Introspective Style Attribution using Diffusion Featurescited by 3
  7. 2025Learning to Weight Parameters for Training Data Attributioncited by 3
  8. 2025Concept-TRAK: Understanding how diffusion models learn concepts through concept-level attributioncited by 3
  9. 2025Attribution-by-design: Ensuring Inference-Time Provenance in Generative Music Systemscited by 3
  10. 2026Deep Sketch-Based 3D Modeling: A Surveycited by 3
  11. 2025The Iconicity of the Generated Imagecited by 2
  12. 2025Identifying Prompted Artist Names from Generated Imagescited by 1
10 more
  1. 2025Coffee: Controllable Diffusion Fine-tuningcited by 1
  2. 2026Unsupervised Synthetic Image Attribution: Alignment and Disentanglementcited by 1
  3. 2026Training data attribution in diffusion models via mirrored unlearning and noise-consistent skewcited by 1
  4. 2026GUDA: Counterfactual Group-wise Training Data Attribution for Diffusion Models via Unlearningcited by 0
  5. 2026ROKA: Robust Knowledge Unlearning against Adversariescited by 0
  6. 2026On the Fragility of Data Attribution When Learning Is Distributedcited by 0
  7. 2026Data Attribution in Large Language Models via Bidirectional Gradient Optimizationcited by 0
  8. 2026FrED: External Data Influence Estimation via Domain Knowledge Graph Groundingcited by 0
  9. 2026Distilling Diffusion Score Discrepancy for Efficient Training Data Attributioncited by 0
  10. 2026TrueMuse: A Benchmark for Data Attribution in Text-to-Music Modelscited by 0