AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

Progressive Differentiable Architecture Search: Bridging the Depth Gap Between Search and Evaluation

Xin Chen, Lingxi Xie, Jun Wu, Qi Tian

ICCV 2019 · Architectures and efficiency › Neural architecture search · approach: Bi-level differentiable NAS · 733 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 operation-level Dropout on skip-connect (M=2) (ablation)P-DARTS search · CIFAR10test error (%): 2.93 -> 2.69Improved
add operation-level Dropout on skip-connect (M=4) (ablation)P-DARTS search · CIFAR10test error (%): 3.51 -> 2.97Improved
add architecture refinement fixing skip-connects at 2 (ablation)P-DARTS search · CIFAR10test error (%): 2.79 -> 2.65Improved
approximate search space from earlier stages instead of random sampling (ablation)P-DARTS final stage · CIFAR10test error (%): 3.43 -> 2.58Improved
remove all skip-connects (M=0) versus M=1 (ablation)P-DARTS cell · CIFAR10test error (%): 2.68 -> 2.78Hurt
increase kept skip-connects from 2 to 4 (ablation)P-DARTS cell · CIFAR10test error (%): 2.69 -> 2.97Hurt
search at 17 cells in final stage instead of DARTS 8-cell searchP-DARTS · CIFAR10test error (%): 2.76 -> 2.50Improved
use P-DARTS instead of DARTS second order on ImageNetP-DARTS · ImageNettop-1 error (%): 26.7 -> 24.4Improved
search on CIFAR100 instead of CIFAR10 and evaluate on CIFAR100P-DARTS · CIFAR100test error (%): 16.55 -> 15.92Improved
search on CIFAR10 and evaluate on CIFAR10 instead of CIFAR100-searchedP-DARTS · CIFAR10test error (%): 2.62 -> 2.50Improved
initialize each stage from previous-stage weights (ablation)P-DARTS · CIFAR10accuracyHurt
enlarge evaluation network channels from 36 to 64P-DARTS · CIFAR10test error (%): 2.50 -> 2.25Improved

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