mlsanityData to 13 Oct 2026

LAMP: Extracting Text from Gradients with Language Model Priors

arXiv 2202.08827 ↗First posted 17 Feb 2022cs.LG cs.DCcited by 54

Not on the map. Why?

Abstract (arXiv)

Recent work shows that sensitive user data can be reconstructed from gradient updates, breaking the key privacy promise of federated learning. While success was demonstrated primarily on image data, these methods do not directly transfer to other domains such as text. In this work, we propose LAMP, a novel attack tailored to textual data, that successfully reconstructs original text from gradients. Our attack is based on two key insights: (i) modeling prior text probability with an auxiliary language model, guiding the search towards more natural text, and (ii) alternating continuous and discrete optimization, which minimizes reconstruction loss on embeddings, while avoiding local minima by applying discrete text transformations. Our experiments demonstrate that LAMP is significantly more effective than prior work: it reconstructs 5x more bigrams and 23% longer subsequences on average. Moreover, we are the first to recover inputs from batch sizes larger than 1 for textual models. These findings indicate that gradient updates of models operating on textual data leak more information than previously thought.

Key facts. Cited by 54 papers in the map (data through 13 Oct 2026). Builds on 31 of its 53 references that are in the map. First posted on arXiv 17 Feb 2022.

Builds on

31 of 53 references.

References · 31

  1. 2017Attention Is All You Needcited by 60,106
  2. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,304
  3. 2014Adam: A Method for Stochastic Optimizationcited by 37,093
  4. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,223
  5. 2016Communication-Efficient Learning of Deep Networks from Decentralized Datacited by 7,653
  6. 2018GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understandingcited by 4,880
  7. 2019Advances and Open Problems in Federated Learningcited by 3,040
  8. 2016Deep Learning with Differential Privacycited by 2,718
  9. 2016Federated Learning: Strategies for Improving Communication Efficiencycited by 1,470
  10. 2019TinyBERT: Distilling BERT for Natural Language Understandingcited by 1,157
  11. 2015Aligning Books and Movies: Towards Story-like Visual Explanations by Watching Movies and Reading Bookscited by 1,042
  12. 2019Deep Leakage from Gradientscited by 1,017
19 more
  1. 2018Neural Network Acceptability Judgmentscited by 957
  2. 2020Inverting Gradients -- How easy is it to break privacy in federated learning?cited by 711
  3. 2016Federated Optimization: Distributed Machine Learning for On-Device Intelligencecited by 563
  4. 2020iDLG: Improved Deep Leakage from Gradientscited by 380
  5. 2021See through Gradients: Image Batch Recovery via GradInversioncited by 326
  6. 2021Gradient-based Adversarial Attacks against Text Transformerscited by 234
  7. 2021Evaluating Gradient Inversion Attacks and Defenses in Federated Learningcited by 196
  8. 2020Deep Learning Based Text Classification: A Comprehensive Reviewcited by 141
  9. 2019Federated Learning for Emoji Prediction in a Mobile Keyboardcited by 118
  10. 2021Gradient Inversion with Generative Image Priorcited by 101
  11. 2020R-GAP: Recursive Gradient Attack on Privacycited by 97
  12. 2020Provable Defense against Privacy Leakage in Federated Learning from Representation Perspectivecited by 94
  13. 2020A Framework for Evaluating Gradient Leakage Attacks in Federated Learningcited by 64
  14. 2021TAG: Gradient Attack on Transformer-based Language Modelscited by 63
  15. 2022Decepticons: Corrupted Transformers Breach Privacy in Federated Learning for Language Modelscited by 53
  16. 2021Towards General Deep Leakage in Federated Learningcited by 40
  17. 2021Bayesian Framework for Gradient Leakagecited by 34
  18. 2021PRECODE - A Generic Model Extension to Prevent Deep Gradient Leakagecited by 28
  19. 2021APRIL: Finding the Achilles' Heel on Privacy for Vision Transformerscited by 24

Built on by

Cited by year (2026 to 13 Oct)

Citing papers per year: 2022: 5; 2023: 11; 2024: 17; 2025: 8; 2026: 13.

54 papers.

Sort:

By area:

  1. 2023Towards Building the Federated GPT: Federated Instruction Tuningcited by 185
  2. 2024Security and Privacy Challenges of Large Language Models: A Surveycited by 185
  3. 2022Recovering Private Text in Federated Learning of Language Modelscited by 70
  4. 2023Locally Differentially Private Document Generation Using Zero Shot Promptingcited by 63
  5. 2023Privacy in Large Language Models: Attacks, Defenses and Future Directionscited by 50
  6. 2023Privacy Implications of Retrieval-Based Language Modelscited by 32
  7. 2023Integration of Large Language Models and Federated Learningcited by 30
  8. 2023Identifying and Mitigating Privacy Risks Stemming from Language Models: A Surveycited by 30
  9. 2022Data Leakage in Federated Averagingcited by 26
  10. 2023FwdLLM: Efficient FedLLM using Forward Gradientcited by 25
  11. 2023Grounding Foundation Models through Federated Transfer Learning: A General Frameworkcited by 24
  12. 2024Deconstructing The Ethics of Large Language Models from Long-standing Issues to New-emerging Dilemmas: A Surveycited by 23
38 more
  1. 2023A Survey of What to Share in Federated Learning: Perspectives on Model Utility, Privacy Leakage, and Communication Efficiencycited by 21
  2. 2024AI Risk Management Should Incorporate Both Safety and Securitycited by 17
  3. 2022Federated Few-Shot Learning for Mobile NLPcited by 15
  4. 2024SPEAR:Exact Gradient Inversion of Batches in Federated Learningcited by 15
  5. 2024DAGER: Exact Gradient Inversion for Large Language Modelscited by 14
  6. 2022TabLeak: Tabular Data Leakage in Federated Learningcited by 13
  7. 2024Trustworthy Distributed AI Systems: Robustness, Privacy, and Governancecited by 13
  8. 2024Inf2Guard: An Information-Theoretic Framework for Learning Privacy-Preserving Representations against Inference Attackscited by 13
  9. 2024Textual Unlearning Gives a False Sense of Unlearningcited by 13
  10. 2024Information Leakage from Embedding in Large Language Modelscited by 10
  11. 2023Hiding in Plain Sight: Disguising Data Stealing Attacks in Federated Learningcited by 9
  12. 2025LLM Security: Vulnerabilities, Attacks, Defenses, and Countermeasurescited by 8
  13. 2025Uncovering Gradient Inversion Risks in Practical Language Model Trainingcited by 7
  14. 2023Beyond Gradient and Priors in Privacy Attacks: Leveraging Pooler Layer Inputs of Language Models in Federated Learningcited by 6
  15. 2025GRAIN: Exact Graph Reconstruction from Gradientscited by 5
  16. 2024FedEval-LLM: Federated Evaluation of Large Language Models on Downstream Tasks with Collective Wisdomcited by 3
  17. 2024Seeing the Forest through the Trees: Data Leakage from Partial Transformer Gradientscited by 3
  18. 2024MIP: CLIP-based Image Reconstruction from PEFT Gradientscited by 2
  19. 2024Privacy in LLM-based Recommendation: Recent Advances and Future Directionscited by 2
  20. 2024Privacy-Preserving Data Deduplication for Enhancing Federated Learning of Language Models (Extended Version)cited by 2
  21. 2025SVDefense: Effective Defense against Gradient Inversion Attacks via Singular Value Decompositioncited by 2
  22. 2026Reverse-Engineering Model Editing on Language Modelscited by 2
  23. 2026Mitigating Gradient Inversion Risks in Language Models via Token Obfuscationcited by 2
  24. 2022Swing Distillation: A Privacy-Preserving Knowledge Distillation Frameworkcited by 1
  25. 2025Empirical Calibration and Metric Differential Privacy in Language Modelscited by 1
  26. 2025Hear No Evil: Detecting Gradient Leakage by Malicious Servers in Federated Learningcited by 1
  27. 2026The Art of Mixology: Mixup-based Obfuscation for Privacy-Preserving Split Learning in Large Language Modelscited by 1
  28. 2026Privacy Failure in Split-LLM Training, The Returned Gradient Nullifies the Decoyscited by 1
  29. 2026Privacy-Preserving Split Learning for Federated LLM Fine-Tuningcited by 1
  30. 2024Amplifying Training Data Exposure through Fine-Tuning with Pseudo-Labeled Membershipscited by 0
  31. 2024Subword Embedding from Bytes Gains Privacy without Sacrificing Accuracy and Complexitycited by 0
  32. 2024GDBR: Label Recovery Attack Against Partial Gradient Encryption in Federated Learningcited by 0
  33. 2025Nosy Layers, Noisy Fixes: Tackling DRAs in Federated Learning Systems using Explainable AIcited by 0
  34. 2025Evaluating Selective Encryption Against Gradient Inversion Attackscited by 0
  35. 2026SOMP: Scalable Gradient Inversion for Large Language Models via Subspace-Guided Orthogonal Matching Pursuitcited by 0
  36. 2026FedSpy-LLM: Towards Scalable and Generalizable Data Reconstruction Attacks from Gradients on LLMscited by 0
  37. 2026TIGER: Inverting Transformer Gradients via Embedding-Subspace Distance Optimizationcited by 0
  38. 2026When the Aggregator Cheats: Data-Free Backdoors in Federated LLM-based QA Systemscited by 0

50 most cited of 54 All 54 (JSON)