Research map / Label-efficient and robust learning / Domain adaptation and generalization
Asymmetric Tri-training for Unsupervised 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 |
|---|---|---|---|
| Use asymmetric tri-training instead of source-only or competing adaptation methods | visual domain adaptation · MNIST→SVHN | accuracy: 52.8 vs 40.3 | Helped |
| Remove batch normalization (ablation) | Asymmetric Tri-training · visual domain adaptation | accuracy: 94.0 -> 85.3; 52.8 -> 39.8 | Hurt |
| Remove the weight constraint between F₁ and F₂ (ablation) | Asymmetric Tri-training · MNIST→SVHN | accuracy: 49.7 -> 52.8 | Hurt |
| Remove the weight constraint between F₁ and F₂ (ablation) | Asymmetric Tri-training · MNIST→MNIST-M | accuracy: 94.2 vs 94.0 | No clear effect |
| Stop gradients from F₁,F₂ to shared feature extractor F (ablation) | Asymmetric Tri-training · MNIST→MNIST-M | accuracy: 95.4 vs 94.0 | No clear effect |
| Stop gradients from F₁,F₂ to shared feature extractor F (ablation) | Asymmetric Tri-training · MNIST→SVHN | accuracy: 47.5 -> 52.8 | Hurt |
| Stop gradients from Fₜ to shared feature extractor F (ablation) | Asymmetric Tri-training · MNIST→MNIST-M | accuracy: 56.4 -> 94.0 | Hurt |
| Stop gradients from Fₜ to shared feature extractor F (ablation) | Asymmetric Tri-training · MNIST→SVHN | accuracy: 47.7 vs 52.8 | Helped |
| Stop gradients from both upper branches (ablation) | Asymmetric Tri-training · SYN SIGNS→GTSRB | accuracy: 96.5 vs 96.2 | No clear effect |
| Use asymmetric tri-training rather than the strongest reported baseline | Amazon Reviews · Amazon Reviews (12 adaptation scenarios) | accuracy | Mixed |
Closest papers by meaning
- Unsupervised Domain Adaptation via Structured Prediction Based Selective Pseudo-Labeling (AAAI 2020 · similarity 0.86)
- Cycle Self-Training for Domain Adaptation (NeurIPS 2021 · similarity 0.85)
- Bi-Directional Generation for Unsupervised Domain Adaptation (AAAI 2020 · similarity 0.85)
- Mutual Nearest Neighbor Contrast and Hybrid Prototype Self-Training for Universal Domain Adaptation (AAAI 2022 · similarity 0.84)
- Gradient-Based Sample Selection for Black-Box Universal Domain Adaptation (AAAI 2025 · similarity 0.84)
- Low Category Uncertainty and High Training Potential Instance Learning for Unsupervised Domain Adaptation (AAAI 2024 · similarity 0.83)
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.