AI Paper Atlas

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

Defending Against Universal Perturbations With Shared Adversarial Training

Chaithanya Kumar Mummadi, Thomas Brox, Jan Hendrik Metzen

ICCV 2019 · Adversarial robustness and security › Certified and empirical robustness · approach: Decision-boundary geometry of robustness · 56 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
apply standard adversarial training (s=1) to the ERM modelResNet20 · CIFAR10 universal attackuniversal robustness: 14.9 -> 28.14Improved
raise sharedness from 1 to 64 in adversarial training (ablation)ResNet20 · CIFAR10accuracy at similar robustness: 78.75% -> 89.94%Improved
raise sharedness from 1 to 8 (ablation)ResNet20 · CIFAR10Pareto frontNo clear effect
raise sharedness beyond 64 (ablation)ResNet20 · CIFAR10robustnessNo clear effect
use shared adversarial training with s=64 at no accuracy lossResNet20 · CIFAR10universal robustness: 14.9 -> 22.7Improved
use shared adversarial training instead of defense of Moosavi-Dezfooli et al. (compute)ResNet20 · CIFAR10time per epoch: 3118s -> 189sImproved
use shared adversarial training instead of defense of Perolat et al. (compute)ResNet20 · CIFAR10time per epoch: 3840s -> 189sImproved
use shared adversarial training s=32 at no accuracy lossWRN-50-2-bottleneck · ImageNet 200-class subsetuniversal robustness: 8.4 -> 15.0Improved
use s=32 instead of s=1 at sigma=1.0, eps=10 (ablation)WRN-50-2-bottleneck · ImageNet 200-class subsetaccuracy: 63.51% -> 72.74%Improved
use s=32 instead of s=1 at sigma=1.0, eps=2 (ablation)WRN-50-2-bottleneck · ImageNet 200-class subsetuniversal robustness: 20.17 -> 14.94Hurt
use shared instead of standard adversarial training (ablation)FCN-8 · Cityscapes targeted universal attackuniversal robustness: 62.5 -> 111.7Improved
use shared instead of standard adversarial training (ablation)FCN-8 · Cityscapes untargeted universal attackuniversal robustness: 25 -> 47.8Improved

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