AI Paper Atlas

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

Unlearning Evaluation through Subset Statistical Independence

Chenhao Zhang, Muxing Li, Feng Liu, Weitong Chen, Miao Xu

ICLR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Membership inference and inversion · 2 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
use SDE (split-half HSIC) instead of MMDin/out-of-training detection · CIFAR-10 ResNet-18, R=5%, |S|=400F1: 0.63 -> 0.87Improved
use SDE instead of Wasserstein distancein/out-of-training detection · CIFAR-10 ResNet-18, R=10%, |S|=2000F1: 0.98 -> 0.97No clear effect
increase target subset size from 400 to 2000 (ablation)SDE · CIFAR-10 ResNet-18, R=5%F1: 0.87 -> 0.99Improved
increase forgetting ratio from 5% to 20% (ablation)SDE · SVHN AllCNN, |S|=400F1: SV-AllCNN |S|=400: 0.62 -> 0.50 (R=10%) -> 0.72 (R=20%)Mixed
train retrained model longer (10% -> 40% of epochs) (ablation)SDE · CIFAR-10 ResNet-18F1: 0.69 -> 0.95Improved
evaluate on deeper layers instead of early layers (ablation)SDE · layer-wise, R=10%, |S|=1000F1Improved
choose kernel bandwidth sigma = sqrt(dim) vs extreme values (ablation)SDE · kernel bandwidthF1Improved
use SDE OTR to judge Unroll vs ASR/accuracyUnroll unlearning · CIFAR10-ResNet18, R=10%OTR (%): Retrain 94.00 vs Unroll 4.40Hurt
apply RandLabel unlearningRandLabel · CIFAR10-ResNet18, R=10%OTR (%): 83.20 vs Retrain 94.00Improved

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