Research map / Label-efficient and robust learning / Test-time adaptation
Learning to Generalize across Domains on Single Test Samples
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| condition classifier generation on each target sample (ablation) | Invariant amortized classifier · PACS | mean accuracy: 82.19 -> 83.26 | Helped |
| use hierarchical inference instead of direct concatenation (ablation) | Single sample generalization · PACS | mean accuracy: 83.26 -> 84.15 | Helped |
| add intermediate supervision on the meta-prior (ablation) | Single sample generalization · PACS | mean accuracy: 82.64 -> 84.15 | Helped |
| mimic domain shifts using domain annotations rather than random image splits (ablation) | Single sample generalization · PACS | mean accuracy: 82.07 -> 84.15 | Helped |
| mimic domain shifts using image clusters rather than random image splits (ablation) | Single sample generalization · PACS | mean accuracy: 82.07 -> 83.06 | Helped |
| add meta-learning to episodic training (ablation) | Single sample generalization · PACS | mean accuracy: 82.21 -> 84.15 | Helped |
| increase backbone capacity from AlexNet to ResNet-50 (ablation) | Single sample generalization · PACS | mean accuracy: 72.50 -> 87.51 | Helped |
| increase backbone capacity from AlexNet to ResNet-50 (ablation) | ERM baseline · PACS | mean accuracy: 71.23 -> 84.09 | Helped |
| use single-sample generalization rather than the strongest reported baseline | PACS · PACS, ResNet-50 | mean accuracy: 86.23 -> 87.51 | Helped |
| use single-sample generalization rather than the strongest reported baseline | PACS · PACS, ResNet-18 | mean accuracy: 85.11 -> 84.15 | Hurt |
| use single-sample generalization rather than the strongest reported baseline | Office-Home · Office-Home, ResNet-18 | mean accuracy: 65.63 -> 66.02 | Helped |
| use single-sample generalization rather than the strongest reported baseline | Office-Home · Office-Home, ResNet-50 | mean accuracy: 68.90 -> 71.07 | Helped |
Closest papers by meaning
- Adaptive Methods for Real-World Domain Generalization (CVPR 2021 · similarity 0.92)
- Learning to Generalize: Meta-Learning for Domain Generalization (AAAI 2018 · similarity 0.91)
- Test-Time Domain Adaptation by Learning Domain-Aware Batch Normalization (AAAI 2024 · similarity 0.88)
- Revisiting Source-Free Domain Adaptation: Insights into Representativeness, Generalization, and Variety (CVPR 2025 · similarity 0.88)
- A General Representation-Based Approach to Multi-Source Domain Adaptation (ICML 2025 · similarity 0.88)
- Lost Domain Generalization Is a Natural Consequence of Lack of Training Domains (AAAI 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.