Research map / Label-efficient and robust learning / Domain adaptation and generalization
Domain Generalization using Causal Matching
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 |
|---|---|---|---|
| match representations for the same slab value instead of relying on ERM | Slab Dataset · Slab Dataset target | accuracy: 57.6 (6.58) -> 77.8 (6.01) | Helped |
| use object-based perfect matches rather than random same-class matches | Rotated MNIST · Rotated MNIST, source 30, 45, 60; target 0° and 90° | accuracy: 78.3 (0.55) -> 83.6 (1.44) | Helped |
| use object-based perfect matches rather than random same-class matches | Rotated Fashion-MNIST · Rotated Fashion-MNIST, source 30, 45, 60; target 0° and 90° | accuracy: 38.4 (2.73) -> 43.8 (1.33) | Helped |
| add augmentation-derived perfect matches | MatchDG on Chest X-rays · Chest X-rays, RSNA target | accuracy: 55.1 (2.93) -> 64.3 (0.75) | Helped |
| add augmentation-derived perfect matches | MatchDG on Chest X-rays · Chest X-rays, ChexPert target | accuracy: 59.0 (0.25) -> 60.6 (0.25) | Hurt |
| use matching-based methods | MatchDG and MDGHybrid on Chest X-rays · Chest X-rays across target domains | accuracy: 64.3 (0.75); 60.6 (0.25); 57.6 (0.13) | Mixed |
| learn matches iteratively instead of using random same-class matches (ablation) | MatchDG · Rotated MNIST | accuracy: 93.4 (0.26) -> 95.1 (0.25); 78.3 (0.55) -> 83.6 (1.44); 63.8 (3.92) -> 69.7 (1.30) | Helped |
| learn matches iteratively instead of using random same-class matches (ablation) | MatchDG · Rotated Fashion-MNIST | accuracy: 77.0 (0.42) -> 80.9 (0.26); 38.4 (2.73) -> 43.8 (1.33); 26.9 (0.34) -> 33.0 (0.72) | Helped |
| use a two-phase method rather than optimize classification and matching simultaneously (ablation) | MatchDG · Rotated MNIST and Fashion-MNIST match recovery | top-10 overlap (%) | Helped |
| add perfect-match information to MatchDG (ablation) | MatchDG · PACS average, ResNet-18 | accuracy: 83.21 -> 84.35 | Helped |
| add perfect-match information to MatchDG (ablation) | MatchDG · PACS average, ResNet-50 | accuracy: 86.11 -> 87.52 | Helped |
| increase the fraction of perfect matches (ablation) | simulated random matches · Rotated MNIST and Fashion-MNIST | accuracy | Helped |
Closest papers by meaning
- Causality Inspired Representation Learning for Domain Generalization (CVPR 2022 · similarity 0.90)
- Do causal predictors generalize better to new domains? (NeurIPS 2024 · similarity 0.88)
- Rethinking Multi-domain Generalization with A General Learning Objective (CVPR 2024 · similarity 0.88)
- Robust Domain Generalization under Divergent Marginal and Conditional Distributions (NeurIPS 2026 · similarity 0.87)
- Enhancing Evolving Domain Generalization through Dynamic Latent Representations (AAAI 2024 · similarity 0.87)
- Towards Generalizing to Unseen Domains with Few Labels (CVPR 2024 · similarity 0.87)
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.