Research map / Label-efficient and robust learning / Domain adaptation and generalization
Disentangling Masked Autoencoders for Unsupervised Domain Generalization
An Zhang, Han Wang, Xiang Wang, Tat-Seng Chua
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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace MAE with DisMAE (disentanglement plus adaptive contrastive loss) | MAE · DomainNet UDG, 100% labels | Avg. accuracy: 38.75 -> 44.86 | Improved |
| replace MAE with DisMAE | MAE · DomainNet UDG, 1% labels linear eval | Avg. accuracy: 13.68 -> 15.16 | Improved |
| DisMAE vs best DG baseline | Supervised DG · VLCS | Avg. accuracy: 79.26 -> 80.38 | Improved |
| DisMAE vs best DG baseline | Supervised DG · PACS | Avg. accuracy: 85.87 -> 86.37 | Improved |
| use intra-domain negatives instead of inter-domain negatives in contrastive loss (ablation) | DisMAE · DomainNet ablation | Avg. accuracy: 37.01 -> 39.39 | Improved |
| add adaptive inverse-probability reweighting from domain classifier (ablation) | DisMAE · DomainNet ablation | Avg. accuracy: 38.72 -> 39.39 | Improved |
| use reversed adaptive weights (ablation) | DisMAE · DomainNet ablation | Avg. accuracy: 39.39 -> 36.33 | Hurt |
| use random weights instead of adaptive weights (ablation) | DisMAE · DomainNet ablation | Avg. accuracy: 39.39 -> 38.31 | Hurt |
| increase decoder depth from 1 to 8 layers (ablation) | DisMAE · DomainNet ablation | Avg. accuracy: 39.39 -> 36.59 | Hurt |
| set mask ratio to 80% rather than 50-70% or 90% (ablation) | DisMAE · DomainNet ablation | Avg. accuracy: 37.20 -> 39.39 (vs 90%) | Improved |
Closest papers by meaning
- Minimal Semantic Sufficiency Meets Unsupervised Domain Generalization (NeurIPS 2025 · similarity 0.90)
- Towards Generalizing to Unseen Domains with Few Labels (CVPR 2024 · similarity 0.89)
- Unknown Domain Inconsistency Minimization for Domain Generalization (ICLR 2024 · similarity 0.89)
- START: A Generalized State Space Model with Saliency-Driven Token-Aware Transformation (NeurIPS 2024 · similarity 0.88)
- Multimodal Unsupervised Domain Generalization by Retrieving Across the Modality Gap (ICLR 2025 · similarity 0.88)
- Towards domain-invariant Self-Supervised Learning with Batch Styles Standardization (ICLR 2024 · similarity 0.88)
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.
