Research map / Continual learning / Domain and prompt-based
On Generalizing Beyond Domains in Cross-Domain Continual Learning
Christian Simon, Masoud Faraki, Yi-Hsuan Tsai, Xiang Yu, Samuel Schulter, Yumin Suh, Mehrtash Harandi, Manmohan Chandraker
Check 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 class-specific Mahalanobis metrics and biases in place of a standard classifier | Cross-domain continual learning · DomainNet and OfficeHome, unseen domains | accuracy | Improved |
| Add EMA-based distillation to MSL (ablation) | MSL · DomainNet-Sketch, 10-task | accuracy (%): 54.1 -> 56.4 | Improved |
| Add EMA-based distillation to MSL (ablation) | MSL · PACS and NICO-Animal | accuracy: 2.1% | Improved |
| Use EMA-based model updates for distillation rather than standard knowledge distillation (ablation) | MSL · OfficeHome, 10-task | accuracy: 2% | Improved |
| Use EMA-based distillation (ablation) | MSL · DomainNet, 10-task | backward transfer rate: 10.1% vs 12.8% | Improved |
| Use EMA-based distillation (ablation) | MSL · OfficeHome, 10-task | backward transfer rate: 7.8% vs 12.3% | Improved |
| Increase Mahalanobis metric maximum rank (ablation) | MSL · OfficeHome, 10-task, unseen domains | accuracy: less than 1.5% | No clear effect |
| Increase replay memory size (ablation) | Cross-domain continual learning methods · OfficeHome, 10-task | accuracy: over 5% | Improved |
| Replace Mahalanobis similarity learning with contrastive loss (ablation) | MSL · DomainNet | accuracy (%): 63.2 vs 21.2; 51.3 vs 24.7; 61.8 vs 32.8; 55.6 vs 17.2 | Hurt |
| Replace Mahalanobis similarity learning with center loss (ablation) | MSL · OfficeHome | accuracy (%): 61.6 vs 46.6; 61.4 vs 44.5; 71.7 vs 58.4; 72.7 vs 52.6 | Hurt |
Closest papers by meaning
- Lifelong Domain Adaptation via Consolidated Internal Distribution (NeurIPS 2021 · similarity 0.92)
- Complementary Domain Adaptation and Generalization for Unsupervised Continual Domain Shift Learning (ICCV 2023 · similarity 0.90)
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.89)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.89)
- Make Domain Shift a Catastrophic Forgetting Alleviator in Class-Incremental Learning (AAAI 2025 · similarity 0.88)
- A Unified and General Framework for Continual Learning (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.
