Research map / Label-efficient and robust learning / Domain adaptation and generalization
CLDA: Contrastive Learning for Semi-Supervised Domain Adaptation
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 |
|---|---|---|---|
| combine class-centroid Inter-Domain Contrastive Alignment and instance-level Instance Contrastive Alignment (ablation) | CLDA · Office-Home, Pr→Ar, 3-shot | test accuracy: 68.3 and 67.7 -> 72.4 | Helped |
| add Instance Contrastive Alignment (ablation) | CLDA without Instance Contrastive Alignment · Office-Home, Pr→Ar, 3-shot | test accuracy: 68.3 -> 72.4 | Helped |
| add Inter-Domain Contrastive Alignment (ablation) | CLDA without Inter-Domain Contrastive Alignment · Office-Home, Pr→Ar, 3-shot | test accuracy: 67.7 -> 72.4 | Helped |
| replace Instance Contrastive Alignment with FixMatch (ablation) | CLDA · Office-Home, Pr→Ar, 3-shot | test accuracy: 70.8 vs 72.4 | Hurt |
| replace Instance Contrastive Alignment with L1 consistency loss (ablation) | CLDA · Office-Home, Pr→Ar, 3-shot | test accuracy: 69.4 vs 72.4 | Hurt |
| replace Instance Contrastive Alignment with L2 consistency loss (ablation) | CLDA · Office-Home, Pr→Ar, 3-shot | test accuracy: 69.3 vs 72.4 | Hurt |
| replace model-prediction pseudo-labels with K-means pseudo-labels (ablation) | CLDA · Office-Home, Pr→Ar, 3-shot | test accuracy: 72.4% -> 71.2% | Hurt |
| set α to 1 instead of 4 (ablation) | CLDA · Office-Home, Pr→Ar, 3-shot | test accuracy: 72.4% -> 70.7% | Hurt |
| increase β (ablation) | CLDA · Office-Home, Pr→Ar, 3-shot | validation and test accuracy | Hurt |
| increase the unlabeled-to-labeled minibatch ratio μ up to 4 (ablation) | CLDA · Office-Home, Pr→Ar, 3-shot | test accuracy | Helped |
| increase μ beyond 4 (ablation) | CLDA · Office-Home, Pr→Ar, 3-shot | test accuracy | Hurt |
| use RandAugment with Grayscale instead of horizontal flipping (ablation) | CLDA · Office-Home, Pr→Ar, 3-shot | test accuracy: 68.1 -> 72.4 | Helped |
Closest papers by meaning
- Universal Semi-Supervised Domain Adaptation by Mitigating Common-Class Bias (CVPR 2024 · similarity 0.91)
- Link-based Contrastive Learning for One-Shot Unsupervised Domain Adaptation (CVPR 2025 · similarity 0.89)
- SimProF: A Simple Probabilistic Framework for Unsupervised Domain Adaptation (AAAI 2025 · similarity 0.89)
- Enhancing Semi-supervised Domain Adaptation via Effective Target Labeling (AAAI 2024 · similarity 0.87)
- Invertible Projection and Conditional Alignment for Multi-Source Blended-Target Domain Adaptation (AAAI 2025 · similarity 0.87)
- CASUAL: Conditional Support Alignment for Domain Adaptation with Label Shift (AAAI 2025 · 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.