AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

Cyclic Differentiable Architecture Search

Hongyuan Yu, Houwen Peng, Yan Huang, Jianlong Fu, Hao Du, Liang Wang, Haibin Ling

TPAMI 2022 · Architectures and efficiency › Neural architecture search · approach: Bi-level differentiable NAS

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 joint learning of search and evaluation networks (L_val^{S,E}) (ablation)CDARTS · ImageNettop-1 accuracy: 72.8 -> 75.7Improved
add updating evaluation network weights during joint training (L_val^E) (ablation)CDARTS · ImageNettop-1 accuracy: 75.7 -> 76.6Improved
add l1-norm regularization to alternating search (ablation)DARTS-style alternating optimization · ImageNetsearch stability: 72.8Improved
use 20-cell instead of 8-cell evaluation network (ablation)CDARTS · CIFAR10accuracyImproved
increase search epochs to 90 (ablation)CDARTS · CIFAR10top-1 accuracy: 97.77Improved
scale up to big model (channels 36 to 50)CDARTS · CIFAR10test accuracy: 97.52 -> 98.32Improved
scale up ImageNet big model (channels 48 to 96, 320x320)CDARTS · ImageNettop-1 accuracy: 81.12Improved
CDARTS vs DARTSCDARTS · ImageNettop-1 accuracy: 73.3 -> 76.3Improved
CDARTS vs DARTS V1 on stabilityCDARTS · CIFAR10 five runsmean accuracy: 96.93±0.13 -> 97.52±0.04Improved
rank correlation of architecture weightsCDARTS vs SPOS · NAS rankingKendall tau: 0.19 -> 0.49Improved
CDARTS vs DARTS in NATS-BenchCDARTS · NATS-Bench CIFAR10 testaccuracy: 59.84 -> 94.02Improved
introspective distillation vs DNA-b teacherCDARTS-b · ImageNet chain-structuredtop-1 accuracy: 77.5% -> 78.2%Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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