AI Paper Atlas

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

Explaining Adversarial Robustness of Neural Networks from Clustering Effect Perspective

Yulin Jin, Xiaoyu Zhang, Jian Lou, Xu Ma, Zilong Wang, Xiaofeng Chen

ICCV 2023 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training refinements · 3 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 SAT generalized inner maximization and clustering regularizer to PGD-ATPGD-AT ResNet-18 · CIFAR10, LAFEAT100robust accuracy (%): 37.05 -> 43.08Improved
combine SAT with TRADESTRADES ResNet-18 · CIFAR10, LAFEAT100robust accuracy (%): 42.02 -> 45.35Improved
combine SAT with AWP-ATAWP-AT WRN28x10 · CIFAR10, LAFEAT100robust accuracy (%): 48.61 -> 54.70Improved
combine SAT with AWP-ATAWP-AT ResNet-18 · CIFAR10, PGD100robust accuracy (%): 47.07 -> 48.82Improved
SAT alone vs AWP-AT and TRADES on output-layer attacksSAT ResNet-18 · CIFAR10, PGD100robust accuracy (%): SAT 42.92 vs TRADES 44.88No clear effect
replace TRADES with SAT-TRADES with more generated dataTRADES · CIFAR10 500K generated, AutoAttackrobust accuracy (%): 62.82 -> 62.51No clear effect
run attacks to convergence (1000 iterations)SAT · CIFAR10 ResNet-18robust accuracy (%): LAFEAT100 43.08 -> LAFEAT1000 42.49No clear effect
add clustering regularization L_ClusterSAT training · SVHN ResNet-18convergenceNo 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