AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Towards Practical Control of Singular Values of Convolutional Layers

Alexandra Senderovich, Ekaterina Bulatova, Anton Obukhov, Maxim Rakhuba

NeurIPS 2022 · Architectures and efficiency › CNN design · approach: Compact CNN architectures · 1 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
add TT decomposition rank 256 to SOCLipConvNet-5 · CIFAR-10accuracy: 75.6 ± 0.3 -> 78.3 ± 0.1Improved
add TT decomposition rank 256 to SOCLipConvNet-5 · CIFAR-10 AutoAttackrobust accuracy: 31.1 ± 0.2 -> 34.7 ± 0.6Improved
add TT decomposition rank 256 to SOCLipConvNet-20 · CIFAR-10accuracy: 76.3 ± 0.5 -> 78.4 ± 0.2Improved
use TT rank 128 instead of 256 in SOC-TTLipConvNet-30 · CIFAR-10accuracy: 76.3 ± 1.0 -> 76.0 ± 0.1No clear effect
add TT decomposition rank 128 to SOCLipConvNet-5 · CIFAR-10speedup: 1.0 -> 1.4Improved
add TT decomposition rank 256 to SOCLipConvNet-5 · CIFAR-10speedup: 1.0 -> 0.9Hurt
increase orthogonal loss weight from 5e3 to 5e4 (ablation)SOC-TT rank 256 · CIFAR-10 LipConvNet-5accuracy: 78.37 -> 75.74Hurt
increase orthogonal loss weight from 5e3 to 5e4 (ablation)SOC-TT rank 256 · CIFAR-10 LipConvNet-5Lipschitz constant: 0.93 -> 0.58Improved
clip singular values to 2 instead of 1WRN-16-10 · CIFAR-10accuracy: 95.27 -> 95.99Improved
add TT rank 256 to clipping to 1WRN-16-10 · CIFAR-10accuracy: 95.27 -> 95.12No clear effect
add TT rank 256 to divisionWRN-16-10 · CIFAR-10 AutoAttackrobust accuracy: 53.39 -> 56.28Improved
add TT rank 320 to divisionWRN-16-10 · CIFAR-10accuracy: 95.17 -> 94.93Hurt

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