AI Paper Atlas

Research map / Label-efficient and robust learning / Domain adaptation and generalization

Flatness-Aware Minimization for Domain Generalization

Xingxuan Zhang, Renzhe Xu, Han Yu, Yancheng Dong, Pengfei Tian, Peng Cui

ICCV 2023 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Foundation-model domain generalization · 31 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 AdamW instead of Adam as the default DG optimizer (ablation)ERM (ResNet-50) · Average over 6 DG datasetsaccuracy: 65.6 -> 66.0Improved
use SGD instead of Adam (ablation)ERM (ResNet-50) · PACSaccuracy: 84.2 -> 79.9Hurt
use SGD instead of Adam (ablation)ERM (ResNet-50) · VLCSaccuracy: 77.3 -> 78.1Improved
optimize zeroth- and first-order flatness jointly (FAD) instead of AdamERM (ResNet-50) · Average over 6 DG datasetsaccuracy: 65.6 -> 67.6Improved
use FAD instead of GAM (first-order flatness only) (ablation)ERM (ResNet-50) · Average over 6 DG datasetsaccuracy: 66.6 -> 67.6Improved
use FAD instead of Adam with SWADSWAD · PACSaccuracy: 86.8 -> 88.5Improved
use FAD instead of Adam with SWADSWAD · Average over 5 DG datasetsaccuracy: 65.3 -> 66.5Improved
use FAD instead of Adam with FishrFishr · Average over 5 DG datasetsaccuracy: 64.5 -> 65.8Improved
use FAD instead of Adam with FishrFishr · DomainNetaccuracy: 44.7 -> 43.8Hurt
train longer (15,000 vs 5,000 iterations) with AdamW (ablation)AdamW · Average of PACS, VLCS, OfficeHome, NICO++accuracy: 76.7 -> 76.4No clear effect
train FAD longer (15,000 vs 5,000 iterations) (ablation)FAD · Average of PACS, VLCS, OfficeHome, NICO++accuracy: 78.7 -> 78.9Mixed
train with FAD for only 5,000 iterations versus other optimizers at 15,000FAD · PACS and NICO++accuracyImproved

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