AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Stochastic Downsampling for Cost-Adjustable Inference and Improved Regularization in Convolutional Networks

Jason Kuen, Xiangfei Kong, Zhe Lin, Gang Wang, Jianxiong Yin, Simon See, Yap-Peng Tan

CVPR 2018 · Architectures and efficiency › CNN design · approach: Theory of convolutional inductive bias · 21 citations (OpenAlex)

Read the paper

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 WRN-d40-w4 with SDPointWRN-d40-w4 · CIFAR-10validation error (%): 4.29 -> 3.73Improved
Train WRN-d28-w10 with SDPointWRN-d28-w10 · CIFAR-10validation error (%): 3.84 -> 3.35Improved
Train DenseNetBC-d40-g60 with SDPointDenseNetBC-d40-g60 · CIFAR-10validation error (%): 3.99 -> 3.39Improved
Train WRN-d40-w4 with SDPointWRN-d40-w4 · CIFAR-100validation error (%): 20.78 -> 19.55Improved
Train WRN-d28-w10 with SDPointWRN-d28-w10 · CIFAR-100validation error (%): 18.51 -> 17.53Improved
Train DenseNetBC-d40-g60 with SDPointDenseNetBC-d40-g60 · CIFAR-100validation error (%): 20.00 -> 19.25Improved
Train ResNeXt-d101-c32 with SDPointResNeXt-d101-c32 · ImageNet-1k valTop-1 error (%): 21.2 -> 20.4Improved
Train ResNeXt-d101-c32 with SDPointResNeXt-d101-c32 · ImageNet-1k valTop-5 error (%): 5.6 -> 5.3Improved
Train PreResNet-d101 with SDPointPreResNet-d101 · ImageNet-1k valTop-1 error (%): 22.0 -> 21.4Improved
Train PreResNet-d101 with SDPointPreResNet-d101 · ImageNet-1k valTop-5 error (%): 6.1 -> 5.6Improved
Use instance-specific rather than uniform BatchNorm statistics (ablation)SDPoint · CIFAR cost-error evaluationclassification errorImproved
Add the 0.5 downsampling ratio alongside 0.75 (ablation)ResNeXt-d101-c32 with SDPoint · ImageNet cost-error evaluationclassification errorImproved

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