mlsanityData to 7 Oct 2026

Reconstructing Training Data From Real World Models Trained with Transfer Learning

arXiv 2407.15845 ↗First posted 22 Jul 2024cs.LG cs.AI cs.CRcited by 10

On the map: World models → Video-generation world models and simulators

Abstract (arXiv)

Current methods for reconstructing training data from trained classifiers are restricted to very small models, limited training set sizes, and low-resolution images. Such restrictions hinder their applicability to real-world scenarios. In this paper, we present a novel approach enabling data reconstruction in realistic settings for models trained on high-resolution images. Our method adapts the reconstruction scheme of arXiv:2206.07758 to real-world scenarios -- specifically, targeting models trained via transfer learning over image embeddings of large pre-trained models like DINO-ViT and CLIP. Our work employs data reconstruction in the embedding space rather than in the image space, showcasing its applicability beyond visual data. Moreover, we introduce a novel clustering-based method to identify good reconstructions from thousands of candidates. This significantly improves on previous works that relied on knowledge of the training set to identify good reconstructed images. Our findings shed light on a potential privacy risk for data leakage from models trained using transfer learning.

Key facts. Cited by 10 papers in the map (data through 7 Oct 2026). Placed in Video-generation world models and simulators, one of 11 directions in World models. Builds on 31 of its 46 references that are in the map. First posted on arXiv 22 Jul 2024.

Builds on

31 of 46 references.

Closest ancestors · 2

Same direction, newest first

  1. 2021Learning Transferable Visual Models From Natural Language Supervision26 Feb
  2. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale22 Oct

Other references · 29

  1. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,160
  2. 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,604
  3. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,267
  4. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,029
  5. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,764
  6. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,339
  7. 2021Emerging Properties in Self-Supervised Vision Transformerscited by 6,893
  8. 2022Hierarchical Text-Conditional Image Generation with CLIP Latentscited by 6,539
  9. 2020Extracting Training Data from Large Language Modelscited by 2,144
  10. 2014How transferable are features in deep neural networks?cited by 1,216
  11. 2019A Comprehensive Survey on Transfer Learningcited by 1,154
  12. 2017Deep Image Priorcited by 1,097
17 more
  1. 2019Deep Leakage from Gradientscited by 1,004
  2. 2018The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networkscited by 847
  3. 2023Extracting Training Data from Diffusion Modelscited by 723
  4. 2020Inverting Gradients -- How easy is it to break privacy in federated learning?cited by 707
  5. 2022Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Modelscited by 418
  6. 2018A Survey on Deep Transfer Learningcited by 417
  7. 2023Scalable Extraction of Training Data from (Production) Language Modelscited by 415
  8. 2017Deep Models Under the GAN: Information Leakage from Collaborative Deep Learningcited by 326
  9. 2019Gradient Descent Maximizes the Margin of Homogeneous Neural Networkscited by 308
  10. 2022Splicing ViT Features for Semantic Appearance Transfercited by 209
  11. 2021Evaluating Gradient Inversion Attacks and Defenses in Federated Learningcited by 196
  12. 2020Directional convergence and alignment in deep learningcited by 172
  13. 2022Reconstructing Training Data from Trained Neural Networkscited by 129
  14. 2015Visualizing Deep Convolutional Neural Networks Using Natural Pre-Imagescited by 93
  15. 2022Fishing for User Data in Large-Batch Federated Learning via Gradient Magnificationcited by 72
  16. 2022A Review of Deep Transfer Learning and Recent Advancementscited by 56
  17. 2023Deconstructing Data Reconstruction: Multiclass, Weight Decay and General Lossescited by 24

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 3; 2025: 3; 2026: 4.

10 papers.

  1. 2026The Implicit Bias of Adam and Muon on Smooth Homogeneous Neural Networkscited by 6
  2. 2024Provable Privacy Attacks on Trained Shallow Neural Networkscited by 4
  3. 2025A Law of Data Reconstruction for Random Features (and Beyond)cited by 3
  4. 2026Mitigating Privacy Risk via Forget Set-Free Unlearningcited by 2
  5. 2025ImpMIA: Leveraging Implicit Bias for Membership Inference Attackcited by 1
  6. 2024Slowing Down Forgetting in Continual Learningcited by 0
  7. 2024On the Reconstruction of Training Data from Group Invariant Networkscited by 0
  8. 2025Querying Kernel Methods Suffices for Reconstructing their Training Datacited by 0
  9. 2026Private and interpretable clinical prediction with quantum-inspired tensor train modelscited by 0
  10. 2026A Sharp Transition in Data Reconstruction under Differential Privacycited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2023Real-Fake: Effective Training Data Synthesis Through Distribution Matching
  2. 2025DRAG: Data Reconstruction Attack using Guided Diffusion
  3. 2024Mind the Gap Between Synthetic and Real: Utilizing Transfer Learning to Probe the Boundaries of Stable Diffusion Generated Data
  4. 2025Reconstructing Template-Memorized Images from Natural Prompts
  5. 2021Task2Sim : Towards Effective Pre-training and Transfer from Synthetic Data
  6. 2024Transfer learning with generative models for object detection on limited datasets
  7. 2023Differentially Private Diffusion Models Generate Useful Synthetic Images
  8. 2024Uncovering Hidden Subspaces in Video Diffusion Models Using Re-Identification