AI Paper Atlas

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

ConvNets and ImageNet Beyond Accuracy: Understanding Mistakes and Uncovering Biases

Pierre Stock, Moustapha Cisse

ECCV 2018 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training with feature consistency · 185 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
count a prediction as correct when at least 4 of 5 turkers agree (rectified error) instead of ground-truth label onlyResNet-101 · ImageNet valtop-1 error (%): 22.69 -> 9.47Improved
human rectification of top-1 errorResNet-18 · ImageNet valtop-1 error (%): 31.39 -> 17.93Improved
human rectification of top-1 errorDenseNet-161 · ImageNet valtop-1 error (%): 22.85 -> 10.87Improved
human rectification of top-1 errorDenseNet-121 · ImageNet valtop-1 error (%): 25.53 -> 14.37Improved
human rectification of top-5 error, showing top-1-misclassified images with top-5 predictionsResNet-101 · ImageNet valtop-5 error (%): 6.44 -> 1.94Improved
show top-5-misclassified images instead of top-1-misclassified ones for top-5 rectificationResNet-101 · ImageNet valrectified top-5 error (%): 1.94 -> 1.37Improved
show LIME explanation (top 8 super-pixels) instead of the whole adversarial image to turkersResNet-101 under IFGSM · ImageNet val (20 classes, adversarial)rectified top-1 error (%): 76.78 -> 70.68Improved
show LIME explanation instead of whole imageResNet-101 under IFGSM · ImageNet val (20 classes, adversarial)turker agreement (%): 22.01% to 30.80%Improved
IFGSM adversarial attack (eps 0.001, 10 iterations)ResNet-101 · ImageNet valaccuracy (%): 77.31% down to 7.96%Hurt
use adversarial selection of prototypes and criticisms instead of MMD-criticmodel criticism · Prototype/criticism assignment studysuccess rate (%): 50.64 -> 57.06Improved
use adversarial selection of prototypes instead of MMD-critic (prototypes only)model criticism · Prototype/criticism assignment studysuccess rate (%): 53.18 -> 55.02Improved
add criticisms to prototypes (3 plus 3 instead of 6 prototypes), adversarial selection (ablation)adversarial model criticism · Prototype/criticism assignment studysuccess rate (%): 55.02 -> 57.06Improved

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