mlsanityData to 7 Oct 2026

Learnability and Privacy Vulnerability are Entangled in a Few Critical Weights

arXiv 2603.13186 ↗First posted 13 Mar 2026cs.LG cs.AI cs.CRcited by 0

Not on the map. Why?

Abstract (arXiv)

Prior approaches for membership privacy preservation usually update or retrain all weights in neural networks, which is costly and can lead to unnecessary utility loss or even more serious misalignment in predictions between training data and non-training data. In this work, we observed three insights: i) privacy vulnerability exists in a very small fraction of weights; ii) however, most of those weights also critically impact utility performance; iii) the importance of weights stems from their locations rather than their values. According to these insights, to preserve privacy, we score critical weights, and instead of discarding those neurons, we rewind only the weights for fine-tuning. We show that, through extensive experiments, this mechanism exhibits outperforming resilience in most cases against Membership Inference Attacks while maintaining utility.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 53 of its 66 references that are in the map. First posted on arXiv 13 Mar 2026.

Builds on

53 of 66 references.

References · 53

  1. 2017Attention Is All You Needcited by 59,932
  2. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  3. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  4. 2015Distilling the Knowledge in a Neural Networkcited by 10,076
  5. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  6. 2016Layer Normalizationcited by 5,579
  7. 2016SGDR: Stochastic Gradient Descent with Warm Restartscited by 4,993
  8. 2016Deep Learning with Differential Privacycited by 2,682
  9. 2018Neural Tangent Kernel: Convergence and Generalization in Neural Networkscited by 2,597
  10. 2015Character-level Convolutional Networks for Text Classificationcited by 2,293
  11. 2018The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networkscited by 2,250
  12. 2016Membership Inference Attacks against Machine Learning Modelscited by 2,221
38 more
  1. 2019Machine Unlearningcited by 1,020
  2. 2021Membership Inference Attacks From First Principlescited by 839
  3. 2018SNIP: Single-shot Network Pruning based on Connection Sensitivitycited by 802
  4. 2023A Simple and Effective Pruning Approach for Large Language Modelscited by 776
  5. 2019Linear Mode Connectivity and the Lottery Ticket Hypothesiscited by 627
  6. 2019Importance Estimation for Neural Network Pruningcited by 531
  7. 2018ML-Leaks: Model and Data Independent Membership Inference Attacks and Defenses on Machine Learning Modelscited by 475
  8. 2021Opacus: User-Friendly Differential Privacy Library in PyTorchcited by 375
  9. 2020Amnesiac Machine Learningcited by 305
  10. 2020Systematic Evaluation of Privacy Risks of Machine Learning Modelscited by 273
  11. 2020Comparing Rewinding and Fine-tuning in Neural Network Pruningcited by 238
  12. 2023Model Sparsity Can Simplify Machine Unlearningcited by 206
  13. 2019MemGuard: Defending against Black-Box Membership Inference Attacks via Adversarial Examplescited by 183
  14. 2018Machine Learning with Membership Privacy using Adversarial Regularizationcited by 177
  15. 2022The Privacy Onion Effect: Memorization is Relativecited by 128
  16. 2024Fit and Prune: Fast and Training-free Visual Token Pruning for Multi-modal Large Language Modelscited by 125
  17. 2023Low-Cost High-Power Membership Inference Attackscited by 99
  18. 2019Adversarial Robustness vs Model Compression, or Both?cited by 98
  19. 2020SCOP: Scientific Control for Reliable Neural Network Pruningcited by 92
  20. 2023Learning to Unlearn: Instance-wise Unlearning for Pre-trained Classifierscited by 66
  21. 2021Mitigating Membership Inference Attacks by Self-Distillation Through a Novel Ensemble Architecturecited by 63
  22. 2021Lost in Pruning: The Effects of Pruning Neural Networks beyond Test Accuracycited by 62
  23. 2022Pruning has a disparate impact on model accuracycited by 43
  24. 2022Membership Inference Attacks and Defenses in Neural Network Pruningcited by 42
  25. 2022RelaxLoss: Defending Membership Inference Attacks without Losing Utilitycited by 35
  26. 2023NTK-SAP: Improving neural network pruning by aligning training dynamicscited by 30
  27. 2023Overconfidence is a Dangerous Thing: Mitigating Membership Inference Attacks by Enforcing Less Confident Predictioncited by 26
  28. 2020On the Effectiveness of Regularization Against Membership Inference Attackscited by 22
  29. 2023Gaussian Membership Inference Privacycited by 20
  30. 2023Leave-one-out Distinguishability in Machine Learningcited by 20
  31. 2022Leveraging Adversarial Examples to Quantify Membership Information Leakagecited by 16
  32. 2024Masks, Signs, And Learning Rate Rewindingcited by 15
  33. 2025Does Training with Synthetic Data Truly Protect Privacy?cited by 12
  34. 2023MIST: Defending Against Membership Inference Attacks Through Membership-Invariant Subspace Trainingcited by 11
  35. 2022How Does a Deep Learning Model Architecture Impact Its Privacy? A Comprehensive Study of Privacy Attacks on CNNs and Transformerscited by 8
  36. 2024Mitigating Privacy Risk in Membership Inference by Convex-Concave Losscited by 5
  37. 2025Trustworthy AI: Safety, Bias, and Privacy -- A Surveycited by 5
  38. 2024Center-Based Relaxed Learning Against Membership Inference Attackscited by 4

50 most cited of 53 All 53 (JSON)

Built on by

Not cited yet.