Research map / Federated learning and privacy / Privacy attacks and secure learning
Label Leakage and Protection in Two-party Split Learning
Read the paperCheck 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 |
|---|---|---|---|
| Use gradient norm as the label-scoring signal | Unprotected two-party split learning · Binary classification during training | norm leak AUC: consistently above 0.9 | Helped |
| Use cosine similarity to a positive-example gradient as the label-scoring signal | Unprotected two-party split learning · Binary classification during training | cosine leak AUC: 1 | Helped |
| Add Marvell perturbations with s=4.0 | Split learning gradients · ISIC, cut layer | leak AUC: ≈ 1 -> ≈ 0.5 | Hurt |
| Replace isotropic Gaussian noise with Marvell's optimized class-dependent perturbations | Split learning gradients · ISIC, cut layer; norm and direction attacks | leak AUC: higher than 0.9 -> ≈ 0.6 | Mixed |
| Use isotropic Gaussian noise instead of Marvell's optimized perturbation (ablation) | Split learning gradients · ISIC and Criteo, cut and first layers | leak AUC at matched utility | Hurt |
| Use max_norm instead of isotropic Gaussian noise (ablation) | Split learning gradients · ISIC and Criteo | leak AUC at matched utility | Mixed |
| Increase Marvell's privacy hyperparameter s (ablation) | Marvell · ISIC, cut and first layers | leak AUC: ≈ 1 -> ≈ 0.5 | Helped |
Closest papers by meaning
- IBAS:Imperceptible Backdoor Attacks in Split Learning with Limited Information (AAAI 2025 · similarity 0.87)
- Auditing Privacy Mechanisms via Label Inference Attacks (NeurIPS 2024 · similarity 0.87)
- Enhancing Learning with Label Differential Privacy by Vector Approximation (ICLR 2025 · similarity 0.87)
- A Stealthy Wrongdoer: Feature-Oriented Reconstruction Attack against Split Learning (CVPR 2024 · similarity 0.86)
- Semantic-Aware Multi-Label Adversarial Attacks (CVPR 2024 · similarity 0.85)
- Attack-Aware Noise Calibration for Differential Privacy (NeurIPS 2024 · similarity 0.85)
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.