AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Beyond Dropout: Feature Map Distortion to Regularize Deep Neural Networks

Yehui Tang, Yunhe Wang, Yixing Xu, Boxin Shi, Chao Xu, Chunjing Xu, Chang Xu

AAAI 2020 · Architectures and efficiency › CNN design · approach: Compact CNN architectures · 42 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
Replace RDdrop with feature map distortion on fully connected layersCNN · CIFAR-10accuracy: 83.11 -> 85.24Improved
Replace RDdrop with feature map distortion on fully connected layersCNN · CIFAR-100accuracy: 54.65 -> 56.23Improved
Replace DropBlock with block feature map distortionResNet-56 · CIFAR-10accuracy: 94.18 -> 94.50Improved
Replace DropBlock with block feature map distortionResNet-56 · CIFAR-100accuracy: 73.08 -> 73.71Improved
Replace Dropout with feature map distortionResNet-50 · ImageNet validationtop-1 accuracy: 76.80 to 77.71Improved
Replace DropBlock with block feature map distortionResNet-50 · ImageNet validationtop-1 accuracy: 78.13 $\pm$ 0.05 -> 78.76 $\pm$ 0.05Improved
Use block feature map distortion rather than elementwise feature map distortionResNet-50 · ImageNet validationtop-1 accuracy: 77.71 $\pm$ 0.05 -> 78.76 $\pm$ 0.05Improved
Increase distortion probability or distortion intensityFeature map distortion · CIFAR-100test accuracyMixed
Increase distortion probability or distortion intensityFeature map distortion · CIFAR-100accuracy gap between training and testingImproved

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