Research map / Label-efficient and robust learning / Domain adaptation and generalization
Is user feedback always informative? Retrieval Latent Defending for Semi-Supervised Domain Adaptation without Source Data
Junha Song, Tae Soo Kim, Junha Kim, Gunhee Nam, Thijs Kooi, Jaegul Choo
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 negatively biased feedback instead of random feedback | AdaMatch · DomainNet-126 (ResNet-50) | top-1 accuracy: 67.6 -> 64.5 | Hurt |
| add retrieval latent defending samples to the mini-batch | AdaMatch under NBF · DomainNet-126 (ResNet-50) | top-1 accuracy: 64.5 -> 72.0 | Improved |
| add retrieval latent defending | FixMatch under NBF · DomainNet-126 (ResNet-50) | top-1 accuracy: 63.4 -> 73.2 | Improved |
| add retrieval latent defending | AdaMatch under NBF · DomainNet-126 (ViT-S) | top-1 accuracy: 73.7 -> 75.9 | Improved |
| add retrieval latent defending | Pseudo-labeling under NBF · MIMIC-CXR-V2 PA to AP | AUROC: .7691 -> .7884 | Improved |
| add retrieval latent defending | RIPU under NBF · GTA5 to Cityscapes | mIoU: 54.5 -> 58.0 | Improved |
| use confident-error feedback (NBF-CE) instead of NBF | Pseudo-labeling · MIMIC-CXR-V2 PA to AP | AUROC: .7691 -> .7639 | Hurt |
| select defending samples from the same class as labeled sample (class-aware) (ablation) | RLD with FreeMatch · DomainNet-126 (ResNet-50) | top-1 accuracy: 74.1 -> 74.8 | Improved |
| increase filtering rate p from 0.4 to 0.8 (ablation) | RLD with FreeMatch · DomainNet-126 (ResNet-50) | top-1 accuracy: 74.8 -> 73.7 | Hurt |
| reduce unlabeled samples per batch (1:7 to 1:4 ratio) and add defending samples (ablation) | RLD with FreeMatch · DomainNet-126 (368 feedback) | top-1 accuracy: 74.2 -> 74.8 | Improved |
| add RLD to baseline under ActiveDA feedback from CLUE | AdaMatch · DomainNet-126 (378 feedback) | top-1 accuracy: 68.6 -> 71.5 | Improved |
| add RLD to SemiSDA methods under NBF on ViT-S versus ResNet-50 (smaller gains with ViT) | SemiSDA baselines · DomainNet-126 | top-1 accuracy: FixMatch +9.8 (ResNet-50) vs +2.6 (ViT-S) | Mixed |
Closest papers by meaning
- AdaMatch: A Unified Approach to Semi-Supervised Learning and Domain Adaptation (ICLR 2022 · similarity 0.85)
- Category-Aware Active Domain Adaptation (ICML 2024 · similarity 0.85)
- Enhancing Semi-supervised Domain Adaptation via Effective Target Labeling (AAAI 2024 · similarity 0.85)
- Can We Evaluate Domain Adaptation Models Without Target-Domain Labels? (ICLR 2024 · similarity 0.84)
- Revisiting Source-Free Domain Adaptation: Insights into Representativeness, Generalization, and Variety (CVPR 2025 · similarity 0.84)
- A General Representation-Based Approach to Multi-Source Domain Adaptation (ICML 2025 · similarity 0.84)
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.
