AI Paper Atlas

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

Navigating the Flatlands: Dual Adaptive Sharpness-Aware Minimization for Domain Generalization

Junwen He, Yang He, Lebing Zheng, Zirui Yin, Hong-Yu Zhang, Yulong Wang

ICML 2026 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Foundation-model domain generalization

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 AMDF multi-directional flatness module to SAM (ablation)SAM · DG average (5 datasets)accuracy (%): 64.1 -> 65.1Improved
add DAS domain-adaptive scaling on top of AMDF (ablation)DA-SAM · DG average (5 datasets)accuracy (%): 65.1 -> 66.0Improved
add AMDF to SAM (ablation)SAM · TerraIncaccuracy (%): 45.7 -> 48.7Improved
add DAS on top of AMDF (ablation)DA-SAM · TerraIncaccuracy (%): 48.7 -> 50.0Improved
add DAS on top of AMDF (ablation)DA-SAM · OfficeHomeaccuracy (%): 68.2 -> 68.4No clear effect
use DA-SAM instead of SAMSAM · OfficeHomeaccuracy (%): 68.0 -> 68.4No clear effect
use DA-SAM instead of SAMSAM · DG average (5 datasets)accuracy (%): 64.1 -> 66.0Improved
probe every 30 iterations instead of every iteration (intermittent probing) (ablation)DA-SAM · DG average (5 datasets)accuracy (%): 66.0 -> 66.1Improved
probe every 10 iterations (ablation)DA-SAM · DG average (5 datasets)accuracy (%): 66.0 -> 65.4Hurt
probe every 60 iterations (ablation)DA-SAM · DG average (5 datasets)training time (s/iter): 0.425 -> 0.114Improved
replace Hessian-based AdaHessian with DA-SAM random probingAdaHessian · memory (MB)memory: 6785 -> 4970Improved

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