AI Paper Atlas

Research map / Federated learning and privacy / Differential privacy

Antipodes of Label Differential Privacy: PATE and ALIBI

Mani Malek Esmaeili, Ilya Mironov, Karthik Prasad, Igor Shilov, Florian Tramer

NeurIPS 2021 · Federated learning and privacy › Differential privacy · approach: Private online and user-level learning · 8 citations (OpenAlex)

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
Use ALIBI instead of LP-2ST at eps=3label DP training · CIFAR-100 ResNet18test accuracy: 28.7 -> 55.0Improved
Use ALIBI instead of LP-2ST at eps=4label DP training · CIFAR-100 ResNet18test accuracy: 50.2 -> 65.0Improved
Use ALIBI instead of LP-2ST at eps=5label DP training · CIFAR-100 ResNet18test accuracy: 63.5 -> 69.0Improved
Use ALIBI instead of LP-2ST at eps=8label DP training · CIFAR-100 ResNet18test accuracy: 74.1 -> 74.4No clear effect
Replace Laplace noise with Gaussian noise (AGIBI) (ablation)ALIBI · CIFAR-10 High accuracytheoretical epsilon: 8.0 -> 19Hurt
Replace Laplace noise with Gaussian noise (AGIBI) (ablation)ALIBI · CIFAR-100 High accuracytheoretical epsilon: 6.3 -> 17Hurt
Replace Laplace noise with Gaussian noise (AGIBI) (ablation)ALIBI · CIFAR-100 Medium accuracytheoretical epsilon: 3.0 -> 8.4Hurt
Replace Laplace noise with Gaussian noise (AGIBI) (ablation)ALIBI · CIFAR-10/100empirical privacy loss eps_m: 2.9--4.0 -> 2.1--3.2No clear effect
Insert 1000 canaries in one model vs 10 runs of 100 canaries (ablation)memorization attack heuristic · CIFAR-100 High, ALIBIeps_m 95% CI: 2.8--3.5 -> 2.7--3.4No clear effect
Insert 1000 canaries vs 10x100 canaries (ablation)memorization attack heuristic · CIFAR-10 Medium, ALIBIeps_m 95% CI: 1.0--2.2 -> 0.2--2.5Mixed
Use PATE-FM instead of ALIBIlabel DP training · CIFAR-100 Highempirical privacy loss eps_m: 2.8--3.5 -> 0.4--1.0Improved
Use PATE-FM instead of ALIBIlabel DP training · CIFAR-100 Hightheoretical epsilon: 6.3 -> 715Hurt

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