Research map / Federated learning and privacy / Privacy attacks and secure learning
Soteria: Provable Defense Against Privacy Leakage in Federated Learning From Representation Perspective
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Increase local training epochs from 1 to 10 | Representation inference on CIFAR10 non-IID · CIFAR10 non-IID, B=32 | representation correlation: 0.98 -> 0.70 (FC1); 0.99 -> 0.78 (FC2); 0.99 -> 0.82 (FC3) | Hurt |
| Reduce local batch size from 32 to 16 | Representation inference on CIFAR10 non-IID · CIFAR10 non-IID, E=1 | representation correlation: 0.98 -> 0.82 (FC1); 0.99 -> 0.93 (FC2); 0.99 -> 0.99 (FC3) | Mixed |
| Reduce local batch size from 32 to 8 | Representation inference on CIFAR10 non-IID · CIFAR10 non-IID, E=1 | representation correlation: 0.98 -> 0.85 (FC1); 0.99 -> 0.89 (FC2); 0.99 -> 0.92 (FC3) | Hurt |
| Use IID rather than non-IID data | Representation inference · CIFAR10, E=1, B=32 | representation correlation: 0.98 -> 0.48 (FC1); 0.99 -> 0.31 (FC2); 0.99 -> 0.18 (FC3) | Hurt |
| Use inferred representations instead of convolutional-layer gradients alone for reconstruction | DLG attack · CIFAR10 | reconstruction quality | Helped |
| Perturb the representation at a defended layer while leaving feature-extractor parameters unperturbed | Soteria · CIFAR10 and MNIST, DLG and GS attacks | reconstruction MSE: 160x | Helped |
| Prune representation coordinates selected for large reconstruction sensitivity | Soteria · CIFAR10, GS attack | reconstruction recognizability | Helped |
| Increase pruning in gradient compression to 80% of whole-model parameters | Gradient compression · CIFAR10, GS attack | reconstruction recognizability | Hurt |
| Apply Soteria to FedAvg | FedAvg · MNIST non-IID | convergence | No clear effect |
Closest papers by meaning
- Towards the Robustness of Differentially Private Federated Learning (AAAI 2024 · similarity 0.88)
- Revamping Federated Learning Security from a Defender's Perspective: A Unified Defense with Homomorphic Encrypted Data Space (CVPR 2024 · similarity 0.87)
- Concealing Sensitive Samples against Gradient Leakage in Federated Learning (AAAI 2024 · similarity 0.87)
- Dual Defense: Enhancing Privacy and Mitigating Poisoning Attacks in Federated Learning (NeurIPS 2024 · similarity 0.87)
- Leak and Learn: An Attacker's Cookbook to Train Using Leaked Data from Federated Learning (CVPR 2024 · similarity 0.87)
- MARS: A Malignity-Aware Backdoor Defense in Federated Learning (NeurIPS 2025 · similarity 0.86)
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.