AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

Eliminating Catastrophic Overfitting Via Abnormal Adversarial Examples Regularization

Runqi Lin, Chaojian Yu, Tongliang Liu

NeurIPS 2023 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training refinements · 5 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 AAER regularization to RS-FGSMRS-FGSM · CIFAR10 eps=16/255PGD-50-10 robust accuracy: 0.00 -> 23.87Improved
add AAER regularization to RS-FGSMRS-FGSM · CIFAR10 eps=12/255PGD-50-10 robust accuracy: 0.00 -> 32.17Improved
add AAER regularization to N-FGSMN-FGSM · CIFAR10 eps=32/255PGD-50-10 robust accuracy: 8.30 -> 12.97Improved
add AAER to N-FGSM at small noiseN-FGSM · CIFAR10 eps=8/255PGD-50-10 robust accuracy: 47.91 -> 48.31Improved
add AAER to RS-FGSMRS-FGSM · CIFAR100 eps=8/255PGD-50-10 robust accuracy: 10.58 -> 25.31Improved
use only regularization part (i) number of AAEs (ablation)RS-AAER · CIFAR10 eps=16/255robust accuracy: 23.87 -> 0.00Hurt
remove part (i) number term (ablation)RS-AAER · CIFAR10 eps=16/255robust accuracy: 23.87 -> 12.56Hurt
remove part (ii) prediction-confidence term (ablation)RS-AAER · CIFAR10 eps=16/255natural accuracy: 64.56 -> 56.65Hurt
only use part (iii) logits term (ablation)RS-AAER · CIFAR10 eps=16/255robust accuracy: 23.87 -> 9.77Hurt
enlarge step size alpha (ablation)RS-AAER · CIFAR10 eps=16/255robust vs natural accuracyMixed
increase lambda1 from 1.0 to 2.0 (ablation)AAER · CIFAR10 eps=16/255robust accuracyNo clear effect
AAER training overhead vs FGSMRS/N-FGSM · CIFAR10 training timeseconds per epoch: 11.0 -> 11.2No 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