AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Parameter Prediction for Unseen Deep Architectures

Boris Knyazev, Michal Drozdzal, Graham W. Taylor, Adriana Romero

NeurIPS 2021 · Architectures and efficiency › RNNs and normalization · approach: Efficient network compression and training · 3 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 GatedGNN with an MLP (ablation)GHN-2 · CIFAR-10 ID-testaccuracy: 66.9 -> 42.2Hurt
remove operation-dependent normalization (ablation)GHN-2 · CIFAR-10 ID-testaccuracy: 66.9 -> 62.6Hurt
remove normalization (ablation)GHN-2 · CIFAR-10 OOD-testaccuracy: 56.8 -> 47.1Hurt
remove virtual edges (ablation)GHN-2 · CIFAR-10 ID-testaccuracy: 66.9 -> 61.5Hurt
remove virtual edges (ablation)GHN-2 · CIFAR-10 OOD-testaccuracy: 56.8 -> 53.9Hurt
remove meta-batching (b_m=1) (ablation)GHN-2 · CIFAR-10 ID-testaccuracy: 66.9 -> 54.3Hurt
train with b_m=1 eight times longer (ablation)GHN-2 · CIFAR-10 ID-testaccuracy: 66.9 -> 62.4Hurt
train on 1000 architectures instead of full set (ablation)GHN-2 · CIFAR-10 ID-testaccuracy: 66.9 -> 65.1Hurt
GHN-2 versus GHN-1 on BN-free networksGHN-2 · CIFAR-10 BN-freeaccuracy: 13.7 -> 36.8Improved
GHN-2 versus GHN-1 on ID-testGHN-2 · CIFAR-10 ID-testaccuracy: 51.4 -> 66.9Improved
GHN-2 versus GHN-1 on ImageNet ID-testGHN-2 · ImageNet ID-testtop-5 accuracy: 17.2 -> 27.2Improved
GHN-2 initialization versus He init for ResNet-50ResNet-50 · 100-shot CIFAR-10accuracy: 41.0 -> 56.4Improved

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