AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Robustness Emerges Early in Training Dynamics, but Is Not Preserved

Jiangang Yang, Wenhui Shi, Lu Hu, Jing Xing, Jian Liu

ECCV 2026 · Architectures and efficiency › RNNs and normalization · approach: Representation geometry and collapse · 0 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 EPS (freeze shallow layers early)ResNet-50 · ImageNet-CTop-1 accuracy: 39.2 -> 42.5 (+3.3)Improved
add AWR (revert shallow weights, low LR)ResNet-50 · ImageNet-CTop-1 accuracy: 39.2 -> 43.1 (+3.9)Improved
add AWR with strong augmentationResNet-50 · ImageNet-CTop-1 accuracy: 39.2 -> 46.5 (+7.3)Improved
add EPS or AWR to WideResNet-50WideResNet-50 · ImageNet-100 variantsAvg. mCE: 88.0 -> 77.2 (-10.8) EPS; 88.0 -> 76.5 (-11.5) AWRImproved
add EPS or AWR to MobileViT-SMobileViT-S · ImageNet-100 variantsAvg. mCE: 89.8 -> 89.2 (-0.6) EPS; 89.8 -> 88.5 (-1.3) AWRImproved
add EPS or AWR to EfficientFormer-L1EfficientFormer-L1 · ImageNet-100 variantsAvg. mCE: 71.4 -> 66.7 (-4.7) EPS; 71.4 -> 70.1 (-1.3) AWRImproved
add EPS or AWR to AugMix and AutoAugResNet-50 · ImageNet-1K variantsAvg. mCE: AutoAug 74.9 -> 73.0 (-1.9) AWR; AugMix 73.4 -> 72.4 (-1.0) AWRImproved
add EPS or AWR to CutMix, Label Smoothing, DropoutResNet-50 · ImageNet-1K variantsAvg. mCE: Dropout 77.6 -> 75.7 (-1.9) EPS; clean IN 76.3 -> 73.9Mixed
add framework to YOLOv5 and FasterRCNNYOLOv5 · COCO-CmAP: 19.9 -> 24.3 (+4.4); clean COCO 39.8 -> 37.9Mixed
add framework to DeepLabV3+DeepLabV3+ · Cityscapes-CmIoU: 36.2 -> 39.9 (+3.7)Improved
add EPS or AWR to AdaCont online TTAAdaCont · ImageNet-C severity 5Top-1 accuracy: 34.9 -> 37.5 (+2.6) AWRImproved
add EPS to Tent online TTATent · ImageNet-C severity 5Top-1 accuracy: 37.3 -> 37.4No clear effect

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