AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

Revisiting Parameter Sharing for Automatic Neural Channel Number Search

Jiaxing Wang, Haoli Bai, Jiaxiang Wu, Xupeng Shi, Junzhou Huang, Irwin King, Michael Lyu, Jian Cheng

NeurIPS 2020 · Architectures and efficiency › Neural architecture search · approach: Latency-aware architecture search · 25 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
anneal sharing level (transitionary) instead of fixed ordinal sharing (ablation)APS channel search ResNet-20 · CIFAR-10accuracy: 91.61% -> 92.02%Improved
anneal sharing level (transitionary) instead of fixed independent sharing (ablation)APS channel search ResNet-20 · CIFAR-10accuracy: 91.24% -> 92.02%Improved
transitionary instead of ordinal sharing (ablation)APS channel search ResNet-56 · CIFAR-10accuracy: 92.93% -> 93.42%Improved
transitionary instead of independent sharing (ablation)APS channel search ResNet-56 · CIFAR-10accuracy: 92.85% -> 93.42%Improved
transitionary instead of ordinal sharing (ablation)APS ResNet-18 · ImageNettop-1 accuracy: 68.60% -> 69.34%Improved
transitionary instead of independent sharing (ablation)APS MobileNet-V2 · ImageNettop-1 accuracy: 72.38% -> 72.83%Improved
higher sharing level (ordinal vs independent) speeds early accuracy rise (ablation)APS channel search ResNet-20 · CIFAR-10 warm-uptraining accuracy speedImproved
higher sharing level (ordinal) on architecture discrimination (ablation)APS channel search ResNet-20 · CIFAR-10controller discriminationHurt
transition learning rate tau too large (1e-1) vs 1e-3 (ablation)APS-T · CIFAR-10 ResNet-20accuracy/discrimination: Phi 2.78e4 (1e-1) vs 2.70e2 (1e-3)Hurt
search at 100% FLOPs with APS-T vs original networkResNet-20 · CIFAR-10accuracy: 92.78% -> 93.14%Improved

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