Research map / Federated learning and privacy / Privacy attacks and secure learning
Privacy-Preserving Collaborative Learning With Automatic Transformation Search
Wei Gao, Shangwei Guo, Tianwei Zhang, Han Qiu, Yonggang Wen, Yang Liu
Check your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace untransformed training with the searched hybrid policy | Collaborative ResNet20 training on CIFAR100 · CIFAR100, ResNet20, Adam+Cosine attack | PSNR: 13.88 -> 7.64 | Improved |
| Replace untransformed training with the searched hybrid policy | Collaborative ResNet20 training on CIFAR100 · CIFAR100, ResNet20 | ACC: 76.88 -> 77.92 | Improved |
| Use randomly selected augmentation policies instead of searched policies | Collaborative learning · CIFAR100, ResNet20 | PSNR: 11.41 vs 7.64 | Hurt |
| Apply the hybrid of searched policies rather than one fixed searched policy (ablation) | Collaborative ResNet20 training · CIFAR100, ResNet20 | ACC: 70.56 -> 77.92 | Improved |
| Apply the hybrid of searched policies rather than one fixed searched policy (ablation) | Collaborative ConvNet training · CIFAR100, ConvNet | ACC: 66.98 -> 70.27 | Improved |
| Replace untransformed training with the searched hybrid policy | Collaborative learning · Fashion MNIST, ResNet20 | PSNR: 10.04 -> 7.60 | Improved |
| Replace untransformed training with the searched hybrid policy | Collaborative learning · Fashion MNIST, ConvNet | ACC: 94.25 -> 91.35 | Hurt |
| Use the searched hybrid policy instead of no defense | CIFAR100, ResNet20 collaborative training · Adam+Cosine attack | PSNR: 13.88 -> 7.64 | Improved |
| Use gradient pruning at 99% compression | CIFAR100, ResNet20 collaborative training · Adam+Cosine attack | ACC: 58.33 | Hurt |
| Increase the number of random-search policy candidates from 1500 to 5000 (ablation) | Automatic transformation search · Search candidate-count analysis | privacy score range | No clear effect |
Closest papers by meaning
- Revisiting Gradient Pruning: A Dual Realization for Defending against Gradient Attacks (AAAI 2024 · similarity 0.87)
- High-Fidelity Gradient Inversion in Distributed Learning (AAAI 2024 · similarity 0.86)
- MaSS: Multi-attribute Selective Suppression for Utility-preserving Data Transformation from an Information-theoretic Perspective (ICML 2024 · similarity 0.85)
- Leveraging Per-Instance Privacy for Machine Unlearning (ICML 2025 · similarity 0.84)
- Reconstruction Attacks on Machine Unlearning: Simple Models are Vulnerable (NeurIPS 2024 · similarity 0.84)
- Theoretical Insights in Model Inversion Robustness and Conditional Entropy Maximization for Collaborative Inference Systems (CVPR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
