AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Scattering Networks for Hybrid Representation Learning

Edouard Oyallon, Sergey Zagoruyko, Gabriel Huang, Nikos Komodakis, Simon Lacoste-Julien, Matthew B. Blaschko, Eugene Belilovsky

TPAMI 2018 · Architectures and efficiency › CNN design · approach: Filter design and trained-filter analysis · 86 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 early learned layers with a scattering network plus 1x1 shared local encoderAlexNet-like pipeline · ILSVRC2012 valtop-1 accuracy (%): 56.9 -> 57.0No clear effect
replace early layers of ResNet with scattering, learning at 28x28 resolutionResnet-18 · ILSVRC2012 valtop-5 accuracy (%): 88.8 -> 88.6No clear effect
add scattering front end to wide ResNet in 100-sample regimeWRN 16-8 · CIFAR-10 (100 samples)accuracy (%): 34.7 -> 38.9Improved
add scattering front end to wide ResNet in 500-sample regimeWRN 16-8 · CIFAR-10 (500 samples)accuracy (%): 46.5 -> 54.7Improved
add scattering front end to wide ResNet in 1000-sample regimeWRN 16-8 · CIFAR-10 (1000 samples)accuracy (%): 60.0 -> 62.0Improved
add scattering front end to wide ResNet with full training dataWRN 16-8 · CIFAR-10 (full)accuracy (%): 95.7 -> 93.1Hurt
use scattering hybrid instead of supervised CNN on STL-10 without unlabeled dataCNN · STL-10accuracy (%): 70.1 -> 76.0Improved
use scattering representation instead of random CNN features with linear classifierrandom CNN · ImageNet linear classificationtop-1 accuracy (%): 12.9 -> 17.4Improved
add successive SLE layers F1, F2, F3 on raw scattering for Caltech-101Scattering + linear SVM · Caltech-101accuracy (%): 62.8 -> 82.9Improved
use scattering + FC unsupervised representation instead of DCGAN featuresDCGAN · CIFAR-10 (unsupervised)accuracy (%): 82.8 -> 84.7Improved
replace scattering by a learned 5-layer CNN approximator (ablation)Scat + Cascaded CNN · CIFAR-10accuracy (%): 91.4 -> about 1% lowerHurt

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