mlsanityData to 7 Oct 2026

Learning to Generate Image Embeddings with User-level Differential Privacy

arXiv 2211.10844 ↗First posted 20 Nov 2022cs.LG cs.CR cs.CVcited by 23

Not on the map. Why?

Abstract (arXiv)

Small on-device models have been successfully trained with user-level differential privacy (DP) for next word prediction and image classification tasks in the past. However, existing methods can fail when directly applied to learn embedding models using supervised training data with a large class space. To achieve user-level DP for large image-to-embedding feature extractors, we propose DP-FedEmb, a variant of federated learning algorithms with per-user sensitivity control and noise addition, to train from user-partitioned data centralized in the datacenter. DP-FedEmb combines virtual clients, partial aggregation, private local fine-tuning, and public pretraining to achieve strong privacy utility trade-offs. We apply DP-FedEmb to train image embedding models for faces, landmarks and natural species, and demonstrate its superior utility under same privacy budget on benchmark datasets DigiFace, EMNIST, GLD and iNaturalist. We further illustrate it is possible to achieve strong user-level DP guarantees of ε<4 while controlling the utility drop within 5%, when millions of users can participate in training.

Key facts. Cited by 23 papers in the map (data through 7 Oct 2026). Builds on 61 of its 87 references that are in the map. First posted on arXiv 20 Nov 2022.

Builds on

61 of 87 references.

References · 61

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 11,979
  3. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,441
  4. 2016Communication-Efficient Learning of Deep Networks from Decentralized Datacited by 7,592
  5. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,339
  6. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  7. 2018MobileNetV2: Inverted Residuals and Linear Bottleneckscited by 4,773
  8. 2020Bootstrap your own latent: A new approach to self-supervised Learningcited by 4,712
  9. 2014Deep Learning Face Attributes in the Wildcited by 3,679
  10. 2019Advances and Open Problems in Federated Learningcited by 3,021
  11. 2015FaceNet: A Unified Embedding for Face Recognition and Clusteringcited by 2,833
  12. 2016Deep Learning with Differential Privacycited by 2,682
38 more
  1. 2016Membership Inference Attacks against Machine Learning Modelscited by 2,221
  2. 2020Extracting Training Data from Large Language Modelscited by 2,144
  3. 2021Barlow Twins: Self-Supervised Learning via Redundancy Reductioncited by 1,818
  4. 2018Group Normalizationcited by 1,514
  5. 2020Adaptive Federated Optimizationcited by 1,036
  6. 2018The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networkscited by 847
  7. 2021Membership Inference Attacks From First Principlescited by 839
  8. 2022Fine-Tuning can Distort Pretrained Features and Underperform Out-of-Distributioncited by 722
  9. 2017Renyi Differential Privacycited by 643
  10. 2017Differentially Private Federated Learning: A Client Level Perspectivecited by 451
  11. 2021Large Language Models Can Be Strong Differentially Private Learnerscited by 405
  12. 2021Differentially Private Fine-tuning of Language Modelscited by 392
  13. 2016Concentrated Differential Privacy: Simplifications, Extensions, and Lower Boundscited by 339
  14. 2020A Metric Learning Reality Checkcited by 317
  15. 2021A Field Guide to Federated Optimizationcited by 312
  16. 2020Google Landmarks Dataset v2 -- A Large-Scale Benchmark for Instance-Level Recognition and Retrievalcited by 305
  17. 2013The Composition Theorem for Differential Privacycited by 290
  18. 2019Differentially Private Learning with Adaptive Clippingcited by 218
  19. 2022Unlocking High-Accuracy Differentially Private Image Classification through Scalecited by 218
  20. 2021Practical and Private (Deep) Learning without Sampling or Shufflingcited by 169
  21. 2020Is Local SGD Better than Minibatch SGD?cited by 151
  22. 2017Collecting Telemetry Data Privatelycited by 147
  23. 2020The Discrete Gaussian for Differential Privacycited by 134
  24. 2020Federated Visual Classification with Real-World Data Distributioncited by 126
  25. 2021Large-Scale Differentially Private BERTcited by 117
  26. 2021Hyperparameter Tuning with Renyi Differential Privacycited by 104
  27. 2020Training BatchNorm and Only BatchNorm: On the Expressive Power of Random Features in CNNscited by 103
  28. 2019Are All Layers Created Equal?cited by 94
  29. 2022DigiFace-1M: 1 Million Digital Face Images for Face Recognitioncited by 92
  30. 2022Toward Training at ImageNet Scale with Differential Privacycited by 87
  31. 2018A General Approach to Adding Differential Privacy to Iterative Training Procedurescited by 84
  32. 2014On Using Very Large Target Vocabulary for Neural Machine Translationcited by 80
  33. 2021Federated Reconstruction: Partially Local Federated Learningcited by 78
  34. 2021Learning with User-Level Privacycited by 72
  35. 2020Training Production Language Models without Memorizing User Datacited by 68
  36. 2020Federated Learning with Only Positive Labelscited by 55
  37. 2022FedAvg with Fine Tuning: Local Updates Lead to Representation Learningcited by 55
  38. 2022Connect the Dots: Tighter Discrete Approximations of Privacy Loss Distributionscited by 50

50 most cited of 61 All 61 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 11; 2024: 7; 2025: 4; 2026: 1.

23 papers.

Sort:

By area:

  1. 2023How to DP-fy ML: A Practical Guide to Machine Learning with Differential Privacycited by 150
  2. 2023Differentially Private Diffusion Models Generate Useful Synthetic Imagescited by 85
  3. 2023Federated Learning of Gboard Language Models with Differential Privacycited by 64
  4. 2024Privacy-Preserving Instructions for Aligning Large Language Modelscited by 34
  5. 2023Can Public Large Language Models Help Private Cross-device Federated Learning?cited by 30
  6. 2024PrE-Text: Training Language Models on Private Federated Data in the Age of LLMscited by 29
  7. 2024Fine-Tuning Large Language Models with User-Level Differential Privacycited by 28
  8. 2024Differentially Private Next-Token Prediction of Large Language Modelscited by 21
  9. 2024Mind the Privacy Unit! User-Level Differential Privacy for Language Model Fine-Tuningcited by 21
  10. 2024Threats and Defenses in Federated Learning Life Cycle: A Comprehensive Survey and Challengescited by 14
  11. 2025POPri: Private Federated Learning using Preference-Optimized Synthetic Datacited by 14
  12. 2023Share Your Representation Only: Guaranteed Improvement of the Privacy-Utility Tradeoff in Federated Learningcited by 13
11 more
  1. 2023On the Efficacy of Differentially Private Few-shot Image Classificationcited by 12
  2. 2024A Hassle-free Algorithm for Private Learning in Practice: Don't Use Tree Aggregation, Use BLTscited by 9
  3. 2023The Fair Value of Data Under Heterogeneous Privacy Constraints in Federated Learningcited by 5
  4. 2023Private Matrix Factorization with Public Item Featurescited by 3
  5. 2023Federated Automatic Differentiationcited by 2
  6. 2023An Empirical Evaluation of Federated Contextual Bandit Algorithmscited by 2
  7. 2023Differentially Private Optimization for Non-Decomposable Objective Functionscited by 0
  8. 2025Differential Privacy Personalized Federated Learning Based on Dynamically Sparsified Client Updatescited by 0
  9. 2025Differentially Private 2D Human Pose Estimationcited by 0
  10. 2025Adaptive Prefiltering for High-Dimensional Similarity Search: A Frequency-Aware Approachcited by 0
  11. 2026Scalable and Private Federated Learning Using Distributed Differential Privacy and Secure Aggregationcited by 0