Research map / Federated learning and privacy / Differential privacy
Decouple-and-Sample: Protecting Sensitive Information in Task Agnostic Data Release
Abhishek Singh, Ethan Garza, Ayush Chopra, Praneeth Vepakomma, Vivek Sharma, Ramesh Raskar
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 Suppression with DP-Sampling in local sampling (ablation) | Sanitizer · E5 UTKFace | utility (CAS): 0.208 -> 0.521 | Improved |
| replace Suppression with DP-Sampling in local sampling (ablation) | Sanitizer · E5 CelebA | utility (CAS): 0.7042 -> 0.817 | Improved |
| replace DP-Obfuscation (Laplace noise) with DP-Sampling (ablation) | Sanitizer · E5 UTKFace | utility (CAS): 0.208 -> 0.521 | Improved |
| replace DP-Obfuscation with DP-Sampling (ablation) | Sanitizer · E5 CelebA | leakage: 0.7129 -> 0.6955 | Improved |
| Sanitizer vs best baseline on privacy-utility AuC | Sanitizer · UTKFace-G | AuC: 0.439 -> 0.487 | Improved |
| Sanitizer vs TIPRDC on privacy-utility AuC | Sanitizer · Fairface-R | AuC: 0.441 -> 0.476 | Improved |
| Sanitizer vs Adversarial Noise on CAS utility | Sanitizer · E4 CelebA | utility (CAS): 0.746 -> 0.9022 | Improved |
| Sanitizer vs Adversarial Noise on CAS leakage | Sanitizer · E4 CelebA | leakage: 0.663 -> 0.6955 | Hurt |
| Sanitizer vs Adversarial Noise on CAS leakage | Sanitizer · E4 UTKFace | leakage: 0.695 -> 0.474 | Improved |
| remove each global-decoupler component (aligner, distance correlation, adversary) (ablation) | global-decoupler · UTKFace | leakage/utility trade-off | Improved |
| use global-decoupler instead of beta-VAE with latent interpolation (ablation) | Sanitizer · UTKFace (E2 setup) | privacy-utility trade-off | Improved |
Closest papers by meaning
- DECO-Bench: Unified Benchmark for Decoupled Task-Agnostic Synthetic Data Release (NeurIPS 2024 · similarity 0.89)
- MaSS: Multi-attribute Selective Suppression for Utility-preserving Data Transformation from an Information-theoretic Perspective (ICML 2024 · similarity 0.88)
- A Certified Unlearning Approach without Access to Source Data (ICML 2025 · similarity 0.86)
- MultiPriv: Benchmarking Individual-Level Privacy Reasoning in Vision-Language Models (ICML 2026 · similarity 0.85)
- PID: Prompt-Independent Data Protection Against Latent Diffusion Models (ICML 2024 · similarity 0.85)
- Balancing Privacy and Performance: A Many-in-One Approach for Image Anonymization (AAAI 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.
