AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Deep Learning without Shortcuts: Shaping the Kernel with Tailored Rectifiers

Guodong Zhang, Aleksandar Botev, James Martens

ICLR 2022 · Architectures and efficiency › RNNs and normalization · approach: Normalization layer theory · 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
Replace EOC with TAT for Leaky ReLU50-layer vanilla network · ImageNet, K-FACtop-1 accuracy: 72.6 -> 74.6Improved
Replace EOC with TAT for Leaky ReLU50-layer vanilla network · ImageNet, SGDtop-1 accuracy: 63.7 -> 71.0Improved
Replace EOC with TAT for Leaky ReLU101-layer vanilla network · ImageNet, K-FACtop-1 accuracy: 71.8 -> 74.2Improved
Replace EOC with TAT for Leaky ReLU101-layer vanilla network · ImageNet, SGDtop-1 accuracy: 41.6 -> 70.0Improved
Add rescaled shortcut branches with weight w=0.8 (ablation)TAT ResNet-50 · ImageNet, K-FAC, 90 epochstop-1 accuracy: 74.6 -> 76.4Improved
Add rescaled shortcut branches with weight w=0.8 (ablation)TAT ResNet-50 · ImageNet, SGD, 90 epochstop-1 accuracy: 71.0 -> 76.0Improved
Add rescaled shortcut branches with weight w=0.8 (ablation)TAT ResNet-101 · ImageNet, K-FAC, 90 epochstop-1 accuracy: 74.2 -> 77.8Improved
Add rescaled shortcut branches with weight w=0.8 (ablation)TAT ResNet-101 · ImageNet, SGD, 90 epochstop-1 accuracy: 70.0 -> 77.3Improved
Replace TReLU with PReLU initialized at 0.0 (ablation)50-layer vanilla network · ImageNet, K-FACtop-1 accuracy: 74.6 -> 72.5Hurt
Replace TReLU with PReLU initialized at 0.25 (ablation)50-layer vanilla network · ImageNet, K-FACtop-1 accuracy: 74.6 -> 73.6Hurt
Train the TAT-initialized negative slope instead of fixing it (ablation)TReLU networks · ImageNettop-1 accuracyNo clear effect
Add batch normalization to shortcut-free networks (ablation)50- and 101-layer vanilla networks · ImageNettop-1 accuracy: 72.6 vs 72.8; 71.8 vs 67.6No clear effect

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