mlsanityData to 7 Oct 2026

Understanding Deep Gradient Leakage via Inversion Influence Functions

arXiv 2309.13016 ↗First posted 22 Sep 2023cs.LG cs.CRcited by 6

Not on the map. Why?

Abstract (arXiv)

Deep Gradient Leakage (DGL) is a highly effective attack that recovers private training images from gradient vectors. This attack casts significant privacy challenges on distributed learning from clients with sensitive data, where clients are required to share gradients. Defending against such attacks requires but lacks an understanding of when and how privacy leakage happens, mostly because of the black-box nature of deep networks. In this paper, we propose a novel Inversion Influence Function (I2F) that establishes a closed-form connection between the recovered images and the private gradients by implicitly solving the DGL problem. Compared to directly solving DGL, I2F is scalable for analyzing deep networks, requiring only oracle access to gradients and Jacobian-vector products. We empirically demonstrate that I2F effectively approximated the DGL generally on different model architectures, datasets, modalities, attack implementations, and perturbation-based defenses. With this novel tool, we provide insights into effective gradient perturbation directions, the unfairness of privacy protection, and privacy-preferred model initialization. Our codes are provided in https://github.com/illidanlab/inversion-influence-function.

Key facts. Cited by 6 papers in the map (data through 7 Oct 2026). Builds on 41 of its 59 references that are in the map. First posted on arXiv 22 Sep 2023.

Builds on

41 of 59 references.

References · 41

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  3. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,029
  4. 2016Communication-Efficient Learning of Deep Networks from Decentralized Datacited by 7,592
  5. 2017mixup: Beyond Empirical Risk Minimizationcited by 5,155
  6. 2015Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classificationcited by 3,590
  7. 2016Pointer Sentinel Mixture Modelscited by 2,992
  8. 2016Deep Learning with Differential Privacycited by 2,682
  9. 2017Understanding Black-box Predictions via Influence Functionscited by 1,714
  10. 2016Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translationcited by 1,471
  11. 2019Deep Leakage from Gradientscited by 1,004
  12. 2020Inverting Gradients -- How easy is it to break privacy in federated learning?cited by 707
29 more
  1. 2020Estimating Training Data Influence by Tracing Gradient Descentcited by 565
  2. 2020iDLG: Improved Deep Leakage from Gradientscited by 367
  3. 2021Evaluating Gradient Inversion Attacks and Defenses in Federated Learningcited by 196
  4. 2020On the Iteration Complexity of Hypergradient Computationcited by 154
  5. 2022If Influence Functions are the Answer, Then What is the Question?cited by 142
  6. 2021Federated Learning for Open Bankingcited by 138
  7. 2022Reconstructing Training Data from Trained Neural Networkscited by 129
  8. 2021Robbing the Fed: Directly Obtaining Private Data in Federated Learning with Modified Modelscited by 106
  9. 2021Gradient Inversion with Generative Image Priorcited by 101
  10. 2020R-GAP: Recursive Gradient Attack on Privacycited by 95
  11. 2020Provable Defense against Privacy Leakage in Federated Learning from Representation Perspectivecited by 94
  12. 2020InstaHide: Instance-hiding Schemes for Private Distributed Learningcited by 90
  13. 2021CAFE: Catastrophic Data Leakage in Vertical Federated Learningcited by 76
  14. 2022Fishing for User Data in Large-Batch Federated Learning via Gradient Magnificationcited by 72
  15. 2021TAG: Gradient Attack on Transformer-based Language Modelscited by 63
  16. 2020Federated Learning for Breast Density Classification: A Real-World Implementationcited by 62
  17. 2022GradViT: Gradient Inversion of Vision Transformerscited by 52
  18. 2022Bounding Training Data Reconstruction in Private (Deep) Learningcited by 43
  19. 2023Bounding Training Data Reconstruction in DP-SGDcited by 42
  20. 2021Gradient-Leakage Resilient Federated Learningcited by 40
  21. 2021Bayesian Framework for Gradient Leakagecited by 34
  22. 2020Rethinking Privacy Preserving Deep Learning: How to Evaluate and Thwart Privacy Attackscited by 32
  23. 2020Privacy-preserving Collaborative Learning with Automatic Transformation Searchcited by 31
  24. 2021Measuring Data Leakage in Machine-Learning Models with Fisher Informationcited by 31
  25. 2021PRECODE - A Generic Model Extension to Prevent Deep Gradient Leakagecited by 28
  26. 2022Reconstructing Training Data from Model Gradient, Provablycited by 27
  27. 2021Understanding Training-Data Leakage from Gradients in Neural Networks for Image Classificationcited by 16
  28. 2020Exploring the Security Boundary of Data Reconstruction via Neuron Exclusivity Analysiscited by 10
  29. 2023A Privacy-Preserving Hybrid Federated Learning Framework for Financial Crime Detectioncited by 10

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 5; 2025: 0; 2026: 1.

6 papers.

  1. 2024Training Data Reconstruction: Privacy due to Uncertainty?cited by 4
  2. 2024Visual Privacy Auditing with Diffusion Modelscited by 1
  3. 2024Bridging Today and the Future of Humanity: AI Safety in 2024 and Beyondcited by 1
  4. 2026A Secure and Private Distributed Bayesian Federated Learning Designcited by 1
  5. 2024R-CONV: An Analytical Approach for Efficient Data Reconstruction via Convolutional Gradientscited by 0
  6. 2024Federated Black-Box Adaptation for Semantic Segmentationcited by 0