AI Paper Atlas

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

Distilling Robust and Non-Robust Features in Adversarial Examples by Information Bottleneck

Junho Kim, Byung-Kwan Lee, Yong Man Ro

NeurIPS 2021 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training with feature consistency · 17 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
use NRF (amplify non-robust gradient) instead of PGDadversarial attack on VGG-16 · CIFAR-10 ADVaccuracy under attack: 44.7 -> 27.4Improved
use NRF instead of AutoAttackadversarial attack on WRN-28-10 · SVHN ADVaccuracy under attack: 55.5 -> 13.4Improved
use NRF instead of CWadversarial attack on VGG-16 · Tiny-ImageNet ADVaccuracy under attack: 12.0 -> 6.7Improved
maximize rather than minimize non-robust gradient norm in attack (ablation)NRF · CIFAR-10 VGG-16accuracy under attack: 67.5 -> 27.4Improved
amplify non-robust gradient rather than robust gradient (ablation)NRF · CIFAR-10 VGG-16accuracy under attack: 35.9 -> 27.4Improved
minimize robust feature gradient (ablation)attack · CIFAR-10 VGG-16accuracy under attack: 74.6 vs clean 79.7No clear effect
select robust channels only vs all intermediate features under PGDadversarially trained VGG-16 · CIFAR-10accuracy: 44.71 -> 99.38Improved
increase information bottleneck beta beyond threshold (ablation)IB feature distillation · CIFAR-10, SVHNrobust vs non-robust feature accuracyMixed

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