AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Overinterpretation reveals image classification model pathologies

Brandon Carter, Siddhartha Jain, Jonas Mueller, David Gifford

NeurIPS 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 8 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
Train and evaluate on 5% backward-selection pixel subsets rather than full imagesResNet18 · CIFAR-10 testaccuracy (%): 95.17 ± 0.21 -> 94.96 ± 0.04No clear effect
Replace backward-selection pixel subsets with random pixel subsets (ablation)ResNet18 trained on 5% subsets · CIFAR-10 testaccuracy (%): 94.96 ± 0.04 -> 51.27 ± 0.82Hurt
Apply input dropout to training and test imagesResNet18 · CIFAR-10-Caccuracy (%): 75.08 ± 0.20 -> 80.35 ± 0.39Improved
Apply input dropout to training and test imagesResNet18 · CIFAR-10 testaccuracy (%): 95.17 ± 0.21 -> 94.15 ± 0.26Hurt
Ensemble five independently trained ResNet18 modelsResNet18 · CIFAR-10 testaccuracy (%): 95.17 ± 0.21 -> 96.07Improved
Train and evaluate on 10% backward-selection pixel subsets rather than full imagesInception v3 · ImageNet validationTop 1 accuracy (%): 77.21 -> 71.37 ± 0.15Hurt
Replace 10% backward-selection subsets with 10% random pixel subsets (ablation)Inception v3 trained on subsets · ImageNet validationTop 1 accuracy (%): 71.37 ± 0.15 -> 64.53 ± 0.16Hurt
Replace 10% backward-selection subsets with 10% random pixel subsets (ablation)pre-trained Inception v3 · ImageNet validationTop 1 accuracy (%): 73.87 -> 0.75 ± 0.02Hurt
Train and evaluate on 10% backward-selection subsets rather than full imagesResNet50 · ImageNet validationTop 1 accuracy (%): 76.13 -> 65.71 ± 0.08Hurt
Replace 10% backward-selection subsets with 10% random pixel subsets (ablation)ResNet50 trained on subsets · ImageNet validationTop 1 accuracy (%): 65.71 ± 0.08 -> 55.70 ± 0.24Hurt
Train and evaluate on 5% backward-selection subsets without data augmentation (ablation)CIFAR-10 classifiers · CIFAR-10 test, 5% BS subsetsaccuracy (%): The paper says training without data augmentation slightly improves accuracy.Improved
Calibrate with temperature scaling (ablation)CIFAR-10 classifiers · CIFAR-10 test, SIS sizeSIS size (% of image): 2.36 ± 1.21 -> 2.94 ± 1.39; 2.53 ± 1.53 -> 3.54 ± 1.94; 2.18 ± 1.37 -> 8.26 ± 2.86Mixed

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