AI Paper Atlas

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

ICML 2026 · Federated learning and privacy › Privacy attacks and secure learning · approach: Membership inference and auditing

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 changedWhereMeasuredEffect
add influence-based initialization (IF-Init) before optimization (ablation)OptiFluence · CIFAR-10TPR@0.1%FPR (%): 38.02 -> 99.60Improved
add influence-based initialization (IF-Init) before optimization (ablation)OptiFluence · MNISTTPR@0.1%FPR (%): 33.53 -> 100.00Improved
initialize from IF-Init instead of random for ReMat+TBPTT (ablation)OptiFluence · CIFAR-10TPR@0.1%FPR (%): 18.8 -> 99.6Improved
add canary optimization on top of influence initialization (ablation)OptiFluence · CIFAR-10TPR@0.1%FPR (%): 31.01 -> 99.60Improved
mislabel the high-influence init sample before optimization (ablation)OptiFluence · CIFAR-10TPR@0.1%FPR (%): 99.60 -> 75.52Hurt
use truncated backprop window K in {2,4} instead of full unrolling (ablation)OptiFluence · MNIST/CIFARTPR@0.1%FPR (%)No clear effect
replace in-distribution canaries with OptiFluence canariesOptiFluence · CIFAR-10TPR@0.1%FPR (%): 24.6 -> 99.6Improved
replace in-distribution canaries with OptiFluence canariesOptiFluence · HAM10KTPR@0.1%FPR (%): 0.5 -> 100.0Improved
replace adversarial canaries with OptiFluence canariesOptiFluence · CIFAR-100TPR@0.1%FPR (%): 72.9 -> 100.0Improved
replace mislabeled canaries with OptiFluence canariesOptiFluence · MNISTTPR@0.1%FPR (%): 92.1 -> 100.0Improved
insert canaries into DP-SGD training with increasing epsilonOptiFluence · CIFAR-10 DP-SGDTPR@0.1%FPR (%): 0.0 -> 6.2Improved
transfer ResNet-9 optimized canaries to larger architecturesOptiFluence · CIFAR-10 transferTPR@0.1%FPR (%): 100% TPR@0.1%FPRNo clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy