AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

Improving Auto-Augment via Augmentation-Wise Weight Sharing

Keyu Tian, Chen Lin, Ming Sun, Luping Zhou, Junjie Yan, Wanli Ouyang

NeurIPS 2020 · Architectures and efficiency › Neural architecture search · approach: AutoML pipeline search · 28 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
use AWS-searched augmentation instead of the baseline preprocessingPyramidNet on CIFAR-10 · CIFAR-10top-1 test error (%): 2.67 -> 1.24Improved
use AWS-searched augmentation instead of the baseline preprocessingWRN on CIFAR-100 · CIFAR-100top-1 test error (%): 18.80 -> 14.16Improved
use AWS-searched augmentation instead of the baseline preprocessingResNet-50 on ImageNet · ImageNettop-1 test error (%): 23.7 -> 20.36Improved
use 4× enlarged batch with AWS (ablation)ResNet-50 with AWS · ImageNettop-1 test error (%): 20.61 ± 0.17 -> 20.36 ± 0.15Improved
use augmentation-trained shared weights rather than weights trained without augmentation (ablation)AWS proxy P_AF · CIFAR-10Pearson correlation: 0.55 -> 0.85Improved
use learned shared weights rather than random initialization (ablation)AWS proxy P_AF · CIFAR-10Pearson correlation: 0.36 -> 0.85Improved
fine-tune shared weights before evaluating the policy (ablation)AWS proxy P_AF · CIFAR-10Pearson correlation: 0.045 -> 0.85Improved
apply augmentation during later rather than early training epochs (ablation)ResNet-18 · CIFAR-10model performanceImproved
use AWS instead of AutoAugment for policy searchaugmentation search on CIFAR-10 · CIFAR-10relative time consuming (times): 60 -> 1.5Improved

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