AI Paper Atlas

Research map / Federated learning and privacy / Privacy attacks and secure learning

Learnability and Privacy Vulnerability are Entangled in a Few Critical Weights

Xingli Fang, Jung-Eun Kim

ICLR 2026 · Federated learning and privacy › Privacy attacks and secure learning · approach: Membership inference and auditing · 0 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
add CWRF on top of HAMPResNet18 · CIFAR-10test accuracy (%): 77.79 -> 81.43Improved
add CWRF on top of CCLResNet18 · CIFAR-10 LiRAattack AUC (%): 83.95 -> 64.82Improved
add CWRF on top of RelaxLossResNet18 · CIFAR-10 LiRAattack AUC (%): 70.51 -> 68.31Improved
add CWRF on top of RelaxLossResNet18 · CIFAR-10 RMIATPR at 0.1% FPR (%): 0.52 -> 1.22Hurt
add CWRF on top of DP-SGDViT · CIFAR-10test accuracy (%): 57.63 -> 60.45Improved
add CWRF on top of RelaxLossViT · CIFAR-10 LiRAattack AUC (%): 73.45 -> 55.88Improved
add CWRF on top of HAMPViT · CIFAR-10 LiRAattack AUC (%): 50.53 -> 50.15No clear effect
add CWRF on top of PASTViT · CIFAR-10test accuracy (%): 54.84 -> 54.66No clear effect
fine-tune privacy-invulnerable weights (A3) instead of privacy-vulnerable weights (A2) after rewinding (ablation)CWRF with RelaxLoss, ResNet18, CIFAR-100 · 3.0% rewoundtest cross-entropy loss: 0.9288 -> 0.8044Improved
rewind privacy-vulnerable weights (A3) instead of removing them (A1) (ablation)CWRF with RelaxLoss, ResNet18, CIFAR-100 · CIFAR-100accuracy recoveryImproved
rewind 0.1% of weights and fine-tune invulnerable ones (A3) versus training from scratch (ablation)RelaxLoss · test cross-entropy losstest cross-entropy loss: 1.5398 -> 0.7720Improved
increase rewound portion from 0.1% to 5.0% (A3) (ablation)CWRF A3 · test cross-entropy losstest cross-entropy loss: 0.7720 -> 0.8330Hurt

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