Research map / Label-efficient and robust learning / Domain adaptation and generalization
What's in a Latent? Leveraging Diffusion Latent Space for Domain Generalization
Xavier Thomas, Deepti Ghadiyaram
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 |
|---|---|---|---|
| augment classifier features with DiT pseudo-domain centroids (ablation) | ERM · DomainBed avg | accuracy: 68.7 -> 70.6 | Improved |
| augment with SD-2.1 pseudo-domain features (ablation) | ERM · DomainBed avg | accuracy: 68.7 -> 71.0 | Improved |
| use CLIP pseudo-domain features (ablation) | ERM · DomainBed avg | accuracy: 68.7 -> 68.4 | No clear effect |
| use DINOv2 pseudo-domain features (ablation) | ERM · DomainBed avg | accuracy: 68.7 -> 69.7 | Improved |
| use SD-2.1 features instead of DiT (ablation) | GUIDE · TerraIncognita | accuracy: 48.2 -> 51.3 | Improved |
| use DiT features instead of SD-2.1 (ablation) | GUIDE · VLCS | accuracy: 77.0 -> 78.5 | Improved |
| add GUIDE pseudo-domain features to ERM | ERM · DomainNet | accuracy: 44.1 -> 45.9 | Improved |
| apply ERM++ training strategies to GUIDE (ablation) | GUIDE · PACS | accuracy: 87.1 -> 89.2 | Improved |
| apply ERM++ training strategies to GUIDE (ablation) | GUIDE · TerraIncognita | accuracy: 51.3 -> 53.6 | Improved |
| add GUIDE to ResNet-50-SWSL backbone | ERM · PACS | accuracy: 86.9 -> 90.0 | Improved |
| add GUIDE to ViT-B/16 backbone | ERM · PACS | accuracy: 88.1 -> 89.7 | Improved |
| add GUIDE to ResNet-18 backbone | ERM · PACS | accuracy: 80.2 -> 81.8 | Improved |
Closest papers by meaning
- Domain Generalization Using a Mixture of Multiple Latent Domains (AAAI 2020 · similarity 0.91)
- Bridging Domain Invariance and Diversity: A Fine-Grained Risk Bound for Domain Generalization (JMLR 2026 · similarity 0.91)
- Adaptive Methods for Real-World Domain Generalization (CVPR 2021 · similarity 0.90)
- Failure Modes of Domain Generalization Algorithms (CVPR 2022 · similarity 0.89)
- Vicinity-Guided Discriminative Latent Diffusion for Privacy-Preserving Domain Adaptation (NeurIPS 2025 · similarity 0.89)
- Weight Diffusion for Future: Learn to Generalize in Non-Stationary Environments (NeurIPS 2024 · similarity 0.89)
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.
