AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

On the Importance of Difficulty Calibration in Membership Inference Attacks

Lauren Watson, Chuan Guo, Graham Cormode, Alexandre Sablayrolles

ICLR 2022 · Adversarial robustness and security › Attacks on text and graphs · approach: Membership inference and inversion · 6 citations (OpenAlex)

Read the paper

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
Apply difficulty calibration to LossScore-based membership inference attacks · Credit, Adult, MNIST, CIFAR10, CIFAR100, ImageNetAUC: 0.542 -> 0.648; 0.516 -> 0.554; 0.505 -> 0.519; 0.676 -> 0.731; 0.911 -> 0.903; 0.547 -> 0.557Mixed
Apply difficulty calibration to Gradient NormScore-based membership inference attacks · Credit, Adult, MNIST, CIFAR10, CIFAR100, ImageNetAUC: 0.513 -> 0.579; 0.510 -> 0.523; 0.504 -> 0.514; 0.678 -> 0.787; 0.912 -> 0.924; 0.540 -> 0.511Mixed
Apply difficulty calibration to ConfidenceScore-based membership inference attacks · Credit, Adult, MNIST, CIFAR10, CIFAR100, ImageNetAUC: 0.501 -> 0.553; 0.507 -> 0.510; 0.503 -> 0.510; 0.629 -> 0.635; 0.852 -> 0.707; 0.544 -> 0.515Mixed
Use difficulty-calibrated scores instead of original scoresScore-based membership inference attacks · Membership inference at low recallprecisionImproved
Use difficulty calibrationScore-based membership inference attacks · Skewed member:non-member ratiosprecisionImproved

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