Research map / Label-efficient and robust learning / Domain adaptation and generalization
Unsupervised Domain Adaptation via Regularized Conditional Alignment
Safa Cicek, Stefano Soatto
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 |
|---|---|---|---|
| add VAT regularizers (ablation) | joint-alignment UDA · MNIST->SVHN | target accuracy (%): 60.65 -> 89.19 | Improved |
| add VAT regularizers (ablation) | joint-alignment UDA · STL->CIFAR | target accuracy (%): 70.20 -> 77.76 | Improved |
| add entropy minimization to the no-VAT model (ablation) | joint-alignment UDA · SVHN->MNIST and MNIST->SVHN | target accuracy (%): SVHN->MNIST 88.33 -> 98.79; MNIST->SVHN 62.95 -> 60.65 | Mixed |
| add source alignment loss (ablation) | joint-alignment UDA · MNIST->SVHN | target accuracy (%): 75.78 -> 89.19 | Improved |
| add source alignment loss (ablation) | joint-alignment UDA · SVHN->MNIST | target accuracy (%): 88.91 -> 99.33 | Improved |
| add target alignment loss (ablation) | joint-alignment UDA · MNIST->SVHN | target accuracy (%): 71.59 -> 89.19 | Improved |
| add both source and target alignment losses (ablation) | joint-alignment UDA · MNIST->SVHN | target accuracy (%): 60.07 -> 89.19 | Improved |
| add both alignment losses (ablation) | joint-alignment UDA · SVHN->MNIST | target accuracy (%): 98.83 -> 99.33 | No clear effect |
| use class predictor instead of joint predictor at inference (ablation) | joint-alignment UDA · MNIST->SVHN | target accuracy (%): 87.88 -> 89.19 | Improved |
| add conditional joint alignment instead of source-only training | UDA · STL->CIFAR | target accuracy (%): 65.44 -> 77.76 | Improved |
| add conditional joint alignment instead of source-only training | UDA · CIFAR->STL | target accuracy (%): 79.41 -> 81.65 | Improved |
| joint (conditional) alignment instead of marginal alignment with input smoothing (VADA+IN) | UDA · MNIST->SVHN | target accuracy (%): 73.3 -> 89.19 | Improved |
Closest papers by meaning
- Co-regularized Alignment for Unsupervised Domain Adaptation (NeurIPS 2018 · similarity 0.93)
- Adversarial Unsupervised Domain Adaptation With Conditional and Label Shift: Infer, Align and Iterate (ICCV 2021 · similarity 0.92)
- Implicit Class-Conditioned Domain Alignment for Unsupervised Domain Adaptation (ICML 2020 · similarity 0.92)
- Unsupervised Domain Adaptation Using Feature-Whitening and Consensus Loss (CVPR 2019 · similarity 0.91)
- CASUAL: Conditional Support Alignment for Domain Adaptation with Label Shift (AAAI 2025 · similarity 0.91)
- Unsupervised Domain Adaptation With Similarity Learning (CVPR 2018 · similarity 0.91)
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.
