Research map / Label-efficient and robust learning / Domain adaptation and generalization
Improving Generalization With Domain Convex Game
Fangrui Lv, Jian Liang, Shuang Li, Jinming Zhang, Di Liu
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 |
|---|---|---|---|
| Use DCG instead of the strongest reported baseline | ResNet-18 DG · PACS leave-one-domain-out | average accuracy: 85.47 -> 86.30 | Improved |
| Use DCG instead of the strongest reported baseline | ResNet-50 DG · PACS leave-one-domain-out | average accuracy: 89.11 -> 89.84 | Improved |
| Use DCG instead of FACT | ResNet-18 DG · Office-Home leave-one-domain-out | average accuracy: 66.56 -> 67.05 | Improved |
| Use DCG instead of CORAL | ResNet-18 DG · mini-DomainNet leave-one-domain-out | average accuracy: 62.87 -> 65.18 | Improved |
| Add Fourier augmentation to original-source supervision (ablation) | PACS baseline · PACS leave-one-domain-out | average accuracy: 79.9 -> 83.4 | Improved |
| Add supermodularity regularization rather than MAML meta-loss (ablation) | Fourier-augmented PACS model · PACS leave-one-domain-out | average accuracy: 84.9 -> 85.6 | Improved |
| Add supermodularity-based sample filtering (ablation) | Fourier-augmented PACS model without regularization · PACS leave-one-domain-out | average accuracy: 83.4 -> 84.4 | Improved |
| Add supermodularity regularization and filtering together rather than use MAML loss and filtering (ablation) | Fourier-augmented PACS model · PACS leave-one-domain-out | average accuracy: 85.7 -> 86.3 | Improved |
| Randomly split augmented domains between meta-train and meta-test (ablation) | DCG meta-domain splitting · PACS leave-one-domain-out | average accuracy: 86.3 -> 85.9 | Hurt |
| Limit sample filtering to augmented samples (ablation) | DCG sample filter · PACS leave-one-domain-out | average accuracy: 86.3 -> 86.1 | Hurt |
| Limit sample filtering to original samples (ablation) | DCG sample filter · PACS leave-one-domain-out | average accuracy: 86.3 -> 86.0 | Hurt |
| Increase domain diversity through augmentation without DCG | Baseline methods · PACS Cartoon and Sketch target domains | accuracy | Mixed |
Closest papers by meaning
- Model-Based Domain Generalization (NeurIPS 2021 · similarity 0.91)
- Domain Generalization: A Survey (TPAMI 2022 · similarity 0.90)
- Gradient-Guided Annealing for Domain Generalization (CVPR 2025 · similarity 0.89)
- Sharpness-Aware Model-Agnostic Long-Tailed Domain Generalization (AAAI 2024 · similarity 0.89)
- Unknown Domain Inconsistency Minimization for Domain Generalization (ICLR 2024 · similarity 0.89)
- One-Step Generalization Ratio Guided Optimization for Domain Generalization (ICML 2025 · 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.
