Research map / Federated learning and privacy / Privacy attacks and secure learning
Concept-Aware Privacy Mechanisms for Defending Embedding Inversion Attacks
Yu-Che Tsai, Hsiang Hsiao, Kuan-Yu Chen, Shou-De Lin
Read the paperCheck 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 spherical Laplace noise with sensitivity-guided Mahalanobis noise | LapMech · STS12 eps=10 | leakage (%): 22.34 -> 19.31 | Improved |
| same, downstream utility | LapMech · STS12 eps=10 | downstream (%): 60.72 -> 65.27 | Improved |
| same, at large budget eps=40 | LapMech · STS12 | downstream (%): 73.98 -> 74.15 | No clear effect |
| use SPARSE with Sentence-T5 embeddings | LapMech · STS12 eps=10 | leakage (%): 31.71 -> 22.38 | Improved |
| use SPARSE against shallow MLC attacker | LapMech · STS12 eps=5 | leakage (%): 19.39 -> 17.63 | Improved |
| SPARSE vs LapMech on STS14 at eps=30 | LapMech · STS14 | leakage (%): 11.22 -> 14.70 | Hurt |
| SPARSE vs LapMech on NFCorpus at eps=5 | LapMech · NFCorpus | downstream (%): 23.70 -> 19.94 | Hurt |
| increase mask regularization (lambda 1e-3 to 1e-2) (ablation) | SPARSE · STS12 eps=10 | leakage (%): 19.31 -> 0.91 | Improved |
| increase mask regularization (lambda 1e-3 to 1e-2) (ablation) | SPARSE · STS12 eps=10 | downstream (%): 65.27 -> 36.87 | Hurt |
| use white-box attribution instead of learned mask (ablation) | SPARSE · STS12 eps=10 | leakage (%): 19.31 -> 12.01 | Improved |
| use SPARSE instead of PII removal | RemovePII · STS12 | downstream (%): 59.47 -> 73.25 | Improved |
Closest papers by meaning
- Split-and-Denoise: Protect large language model inference with local differential privacy (ICML 2024 · similarity 0.88)
- One for All: A Universal Generator for Concept Unlearnability via Multi-Modal Alignment (ICML 2024 · similarity 0.86)
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · similarity 0.86)
- Privacy-Preserving Embedding via Look-up Table Evaluation with Fully Homomorphic Encryption (ICML 2024 · similarity 0.85)
- InvisibleInk: High-Utility and Low-Cost Text Generation with Differential Privacy (NeurIPS 2025 · similarity 0.85)
- Language Models are Advanced Anonymizers (ICLR 2025 · 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.
