AI Paper Atlas

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

Adversarial Examples Detection in Deep Networks With Convolutional Filter Statistics

Xin Li, Fuxin Li

ICCV 2017 · Adversarial robustness and security › Certified and empirical robustness · approach: Regularized defenses against attacks · 388 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 cascade over layers instead of the best single layer (ablation)detector (AlexNet) · Normal vs L-BFGS adversarialsdetection accuracy (%): 78.95 -> 83.4Improved
use cascade over layers instead of the best single layer (ablation)detector (VGG-16) · Normal vs L-BFGS adversarialsdetection accuracy (%): 84.1 -> 90.665Improved
use deeper VGG-16 backbone instead of AlexNet for the cascade (ablation)cascade detector · Normal vs L-BFGS adversarialsdetection accuracy (%): 83.4 -> 90.665Improved
use cascade statistics detector instead of OpenMaxdetector (AlexNet) · Normal vs L-BFGS adversarialsaccuracy / AUC: outperform OpenMax by over 9% in AUC and 11% in accuracyImproved
detect EA adversarials with a detector trained only on L-BFGS adversarialscascade detector (AlexNet) · Normal vs EA adversarialsaccuracy / AUC: 97.3% accuracy, 98.2% AUCImproved
use first three conv layers only for EA detection (ablation)single-layer statistics (AlexNet) · Normal vs EA adversarialsdetection accuracy (%): 93.45 ± 0.69 -> 98.3 ± 0.73Improved
add 3x3 average filter to detected adversarial images before classificationVGG-16 classifier · Recovery of detected adversarialstop-5 accuracy: almost 0% -> 73.0%Improved
use 5x5 instead of 3x3 average filter for recovery (ablation)recovery filter · Recovery of detected adversarialstop-5 accuracy: 73.0% -> 68.0%Hurt
use foveation crop instead of average filter for recoveryrecovery · Recovery of detected adversarialstop-5 accuracy: 73.0% -> 82.6%Improved
use PCA tail-distribution features as detectortail-statistics classifier · Normal vs L-BFGS adversarialsdetection accuracy (%): >99% accuracyMixed
apply detector to DeepFool adversarials on ResNet-50cascade detector · Normal vs DeepFool adversarialsAUC: more than 90% AUCImproved
add first-conv-layer prefiltering decision threshold at 97% (AlexNet) / 98% (VGG) true positive ratecascade detector · Normal vs adversarialstrue positive rate: 97% and 98%No 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