AI Paper Atlas

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

Parametric Scattering Networks

Shanel Gauthier, Benjamin Thérien, Laurent Alsène-Racicot, Muawiz Chaudhary, Irina Rish, Eugene Belilovsky, Michael Eickenberg, Guy Wolf

CVPR 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Dense features and foundation priors · 19 citations (OpenAlex)

Check 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
Learn Morlet wavelet parameters from tight-frame initializationScattering + linear classifier · CIFAR-10, 100 samplesaccuracy: 36.01±0.55 -> 37.84±0.57Improved
Learn Morlet wavelet parameters from tight-frame initializationScattering + WRN · COVIDx CRX-2, 100 samplesaccuracy: 80.50± 1.15 -> 81.20±1.73Improved
Replace Morlet-constrained parameters with pixel-wise wavelet optimization (ablation)Learned scattering + linear classifier · CIFAR-10, 100 samplesaccuracy: 37.84±0.57 -> 32.30 ± 0.69Hurt
Use an equivariant parameterization with shared parameters across orientations (ablation)Learned scattering + linear classifier · CIFAR-10, 100 samplesaccuracy: 37.84±0.57 -> 39.69±0.56No clear effect
Use random rather than tight-frame initialization (ablation)Learned scattering + linear classifier · CIFAR-10, 100 samplesaccuracy: 37.84±0.57 -> 34.81±0.60Hurt
Increase learned filters per scale from 8 to 16 (ablation)Learned scattering + linear classifier · CIFAR-10, 500 samplesaccuracy: 50.74±0.30 -> 54.17±0.36Improved
Increase fixed scattering filters per scale from 8 to 16 (ablation)Fixed scattering + linear classifier · CIFAR-10, 500 samplesaccuracy: 49.67±0.33 -> 48.2 ± 0.27Hurt
Remove Autoaugment (ablation)Learned scattering + linear classifier · CIFAR-10, all training dataaccuracy: 69.57±0.1 -> 74.94±0.06Improved
Remove Autoaugment (ablation)Learned scattering + WRN · CIFAR-10, 100 samplesaccuracy: 43.60±0.87 -> 34.95 ± 0.96Hurt
Train wavelet parameters with SimCLR before linear evaluationScattering representation · CIFAR-10, 100 samplesaccuracy: 36.01±0.55 -> 38.05 ± 0.45Improved
Replace cross-entropy with cosine loss (ablation)Learned scattering + WRN · CIFAR-10accuracy: 43.6±0.87 -> 42.94 ± 0.77Hurt
Replace cross-entropy with cosine loss (ablation)Learned scattering + WRN · KTH-TIPS2, 1188 samplesaccuracy: 69.23±0.67 -> 70.86±0.67Improved

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