mlsanityData to 7 Oct 2026

Do Not Let Privacy Overbill Utility: Gradient Embedding Perturbation for Private Learning

arXiv 2102.12677 ↗First posted 25 Feb 2021cs.LGcited by 86

Not on the map. Why?

Abstract (arXiv)

The privacy leakage of the model about the training data can be bounded in the differential privacy mechanism. However, for meaningful privacy parameters, a differentially private model degrades the utility drastically when the model comprises a large number of trainable parameters. In this paper, we propose an algorithm Gradient Embedding Perturbation (GEP) towards training differentially private deep models with decent accuracy. Specifically, in each gradient descent step, GEP first projects individual private gradient into a non-sensitive anchor subspace, producing a low-dimensional gradient embedding and a small-norm residual gradient. Then, GEP perturbs the low-dimensional embedding and the residual gradient separately according to the privacy budget. Such a decomposition permits a small perturbation variance, which greatly helps to break the dimensional barrier of private learning. With GEP, we achieve decent accuracy with reasonable computational cost and modest privacy guarantee for deep models. Especially, with privacy bound ε=8, we achieve 74.9% test accuracy on CIFAR10 and 95.1% test accuracy on SVHN, significantly improving over existing results.

Key facts. Cited by 86 papers in the map (data through 7 Oct 2026). Builds on 34 of its 61 references that are in the map. First posted on arXiv 25 Feb 2021.

Builds on

34 of 61 references.

References · 34

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2019Momentum Contrast for Unsupervised Visual Representation Learningcited by 7,482
  3. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  4. 2016Deep Learning with Differential Privacycited by 2,682
  5. 2016Membership Inference Attacks against Machine Learning Modelscited by 2,221
  6. 2018Group Normalizationcited by 1,514
  7. 2020Big Self-Supervised Models are Strong Semi-Supervised Learnerscited by 1,375
  8. 2019Deep Leakage from Gradientscited by 1,004
  9. 2018The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networkscited by 847
  10. 2017Renyi Differential Privacycited by 643
  11. 2015An Introduction to Matrix Concentration Inequalitiescited by 391
  12. 2016Semi-supervised Knowledge Transfer for Deep Learning from Private Training Datacited by 347
22 more
  1. 2016Concentrated Differential Privacy: Simplifications, Extensions, and Lower Boundscited by 339
  2. 2017Deep Models Under the GAN: Information Leakage from Collaborative Deep Learningcited by 326
  3. 2018Scalable Private Learning with PATEcited by 266
  4. 2019White-box vs Black-box: Bayes Optimal Strategies for Membership Inferencecited by 265
  5. 2018Privacy Amplification by Subsampling: Tight Analyses via Couplings and Divergencescited by 237
  6. 2018Subsampled R'enyi Differential Privacy and Analytical Moments Accountantcited by 234
  7. 2019PowerSGD: Practical Low-Rank Gradient Compression for Distributed Optimizationcited by 234
  8. 2020Differentially Private Learning Needs Better Features (or Much More Data)cited by 233
  9. 2018Gradient Descent Happens in a Tiny Subspacecited by 208
  10. 2019R'enyi Differential Privacy of the Sampled Gaussian Mechanismcited by 199
  11. 2020Understanding Gradient Clipping in Private SGD: A Geometric Perspectivecited by 140
  12. 2018Differentially Private Empirical Risk Minimization Revisited: Faster and More Generalcited by 127
  13. 2014Differentially Private Empirical Risk Minimization: Efficient Algorithms and Tight Error Boundscited by 104
  14. 2020Bypassing the Ambient Dimension: Private SGD with Gradient Subspace Identificationcited by 88
  15. 2019Data Poisoning against Differentially-Private Learners: Attacks and Defensescited by 75
  16. 2019Limits of Private Learning with Access to Public Datacited by 38
  17. 2019Hessian based analysis of SGD for Deep Nets: Dynamics and Generalizationcited by 37
  18. 2018Concentrated Differentially Private Gradient Descent with Adaptive per-Iteration Privacy Budgetcited by 31
  19. 2020Low-rank Gradient Approximation For Memory-Efficient On-device Training of Deep Neural Networkcited by 20
  20. 2017Differentially Private Empirical Risk Minimization with Input Perturbationcited by 19
  21. 2019Assessing differentially private deep learning with Membership Inferencecited by 11
  22. 2020Fast Dimension Independent Private AdaGrad on Publicly Estimated Subspacescited by 10

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 8; 2022: 20; 2023: 22; 2024: 19; 2025: 12; 2026: 5.

86 papers.

Sort:

By area:

  1. 2021Large Language Models Can Be Strong Differentially Private Learnerscited by 405
  2. 2021Differentially Private Fine-tuning of Language Modelscited by 392
  3. 2022Unlocking High-Accuracy Differentially Private Image Classification through Scalecited by 218
  4. 2022Synthetic Text Generation with Differential Privacy: A Simple and Practical Recipecited by 108
  5. 2021Trustworthy AI: A Computational Perspectivecited by 93
  6. 2021Large Scale Private Learning via Low-rank Reparametrizationcited by 87
  7. 2022Position: Considerations for Differentially Private Learning with Large-Scale Public Pretrainingcited by 86
  8. 2022Mixed Differential Privacy in Computer Visioncited by 53
  9. 2022Exploring the Limits of Differentially Private Deep Learning with Group-wise Clippingcited by 52
  10. 2021Public Data-Assisted Mirror Descent for Private Model Trainingcited by 49
  11. 2022When Does Differentially Private Learning Not Suffer in High Dimensions?cited by 48
  12. 2022Scalable and Efficient Training of Large Convolutional Neural Networks with Differential Privacycited by 47
38 more
  1. 2022Differentially Private Bias-Term Fine-tuning of Foundation Modelscited by 47
  2. 2021Private Adaptive Gradient Methods for Convex Optimizationcited by 46
  3. 2024Private Fine-tuning of Large Language Models with Zeroth-order Optimizationcited by 41
  4. 2022Gradient Obfuscation Gives a False Sense of Security in Federated Learningcited by 37
  5. 2023Why Is Public Pretraining Necessary for Private Model Training?cited by 36
  6. 2022Federated Learning with Sparsified Model Perturbation: Improving Accuracy under Client-Level Differential Privacycited by 32
  7. 2022Private Estimation with Public Datacited by 31
  8. 2021DPNAS: Neural Architecture Search for Deep Learning with Differential Privacycited by 20
  9. 2023Differentially Private Image Classification by Learning Priors from Random Processescited by 19
  10. 2023Unlocking Accuracy and Fairness in Differentially Private Image Classificationcited by 16
  11. 2022Fine-Tuning with Differential Privacy Necessitates an Additional Hyperparameter Searchcited by 15
  12. 2023Choosing Public Datasets for Private Machine Learning via Gradient Subspace Distancecited by 14
  13. 2023Differentially Private Sharpness-Aware Trainingcited by 14
  14. 2023Private Distribution Learning with Public Data: The View from Sample Compressioncited by 14
  15. 2022Differentially Private Model Compressioncited by 13
  16. 2023DPMLBench: Holistic Evaluation of Differentially Private Machine Learningcited by 13
  17. 2022Differential Privacy in Natural Language Processing: The Story So Farcited by 12
  18. 2023PILLAR: How to make semi-private learning more effectivecited by 10
  19. 2023Optimal Differentially Private Model Training with Public Datacited by 9
  20. 2023DPSUR: Accelerating Differentially Private Stochastic Gradient Descent Using Selective Update and Releasecited by 9
  21. 2024Noiseless Privacy-Preserving Decentralized Learningcited by 9
  22. 2021Improving Differentially Private SGD via Randomly Sparsified Gradientscited by 8
  23. 2024LLM-based Privacy Data Augmentation Guided by Knowledge Distillation with a Distribution Tutor for Medical Text Classificationcited by 8
  24. 2022Private, Efficient, and Accurate: Protecting Models Trained by Multi-party Learning with Differential Privacycited by 7
  25. 2022Differentially Private Image Classification from Featurescited by 7
  26. 2024DPAdapter: Improving Differentially Private Deep Learning through Noise Tolerance Pre-trainingcited by 7
  27. 2024Efficiently Computing Similarities to Private Datasetscited by 7
  28. 2023Spectral-DP: Differentially Private Deep Learning through Spectral Perturbation and Filteringcited by 6
  29. 2023Resisting Backdoor Attacks in Federated Learning via Bidirectional Elections and Individual Perspectivecited by 6
  30. 2025DC-SGD: Differentially Private SGD with Dynamic Clipping through Gradient Norm Distribution Estimationcited by 6
  31. 2023Differential Privacy Meets Neural Network Pruningcited by 5
  32. 2023On the Benefits of Public Representations for Private Transfer Learning under Distribution Shiftcited by 5
  33. 2024Public-data Assisted Private Stochastic Optimization: Power and Limitationscited by 5
  34. 2022Differentially Private Deep Learning with ModelMixcited by 4
  35. 2022Privacy in Practice: Private COVID-19 Detection in X-Ray Images (Extended Version)cited by 4
  36. 2023Equivariant Differentially Private Deep Learning: Why DP-SGD Needs Sparser Modelscited by 4
  37. 2023All Rivers Run to the Sea: Private Learning with Asymmetric Flowscited by 4
  38. 2024Low-Cost Privacy-Preserving Decentralized Learningcited by 4

50 most cited of 86 All 86 (JSON)