Research map / Federated learning and privacy / Privacy attacks and secure learning
OptiFluence: Principled Design of Privacy Canaries
Mohammad Yaghini, Michael Aerni, Junrui Zhang, Nicolas Papernot, Florian Tramèr
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 |
|---|---|---|---|
| add influence-based initialization (IF-Init) before optimization (ablation) | OptiFluence · CIFAR-10 | TPR@0.1%FPR (%): 38.02 -> 99.60 | Improved |
| add influence-based initialization (IF-Init) before optimization (ablation) | OptiFluence · MNIST | TPR@0.1%FPR (%): 33.53 -> 100.00 | Improved |
| initialize from IF-Init instead of random for ReMat+TBPTT (ablation) | OptiFluence · CIFAR-10 | TPR@0.1%FPR (%): 18.8 -> 99.6 | Improved |
| add canary optimization on top of influence initialization (ablation) | OptiFluence · CIFAR-10 | TPR@0.1%FPR (%): 31.01 -> 99.60 | Improved |
| mislabel the high-influence init sample before optimization (ablation) | OptiFluence · CIFAR-10 | TPR@0.1%FPR (%): 99.60 -> 75.52 | Hurt |
| use truncated backprop window K in {2,4} instead of full unrolling (ablation) | OptiFluence · MNIST/CIFAR | TPR@0.1%FPR (%) | No clear effect |
| replace in-distribution canaries with OptiFluence canaries | OptiFluence · CIFAR-10 | TPR@0.1%FPR (%): 24.6 -> 99.6 | Improved |
| replace in-distribution canaries with OptiFluence canaries | OptiFluence · HAM10K | TPR@0.1%FPR (%): 0.5 -> 100.0 | Improved |
| replace adversarial canaries with OptiFluence canaries | OptiFluence · CIFAR-100 | TPR@0.1%FPR (%): 72.9 -> 100.0 | Improved |
| replace mislabeled canaries with OptiFluence canaries | OptiFluence · MNIST | TPR@0.1%FPR (%): 92.1 -> 100.0 | Improved |
| insert canaries into DP-SGD training with increasing epsilon | OptiFluence · CIFAR-10 DP-SGD | TPR@0.1%FPR (%): 0.0 -> 6.2 | Improved |
| transfer ResNet-9 optimized canaries to larger architectures | OptiFluence · CIFAR-10 transfer | TPR@0.1%FPR (%): 100% TPR@0.1%FPR | No clear effect |
Closest papers by meaning
- Privacy Auditing of Large Language Models (ICLR 2025 · similarity 0.90)
- The Last Iterate Advantage: Empirical Auditing and Principled Heuristic Analysis of Differentially Private SGD (ICLR 2025 · similarity 0.89)
- Tighter Privacy Auditing of DP-SGD in the Hidden State Threat Model (ICLR 2025 · similarity 0.88)
- One-shot Empirical Privacy Estimation for Federated Learning (ICLR 2024 · similarity 0.87)
- Assessing Per-Sample Membership Inference Vulnerability without Retraining (NeurIPS 2026 · similarity 0.87)
- The Canary’s Echo: Auditing Privacy Risks of LLM-Generated Synthetic Text (ICML 2025 · similarity 0.87)
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.
