Research map / Label-efficient and robust learning / Domain adaptation and generalization
Partially-Shared Variational Auto-encoders for Unsupervised Domain Adaptation with Target Shift
Ryuhei Takahashi, Atsushi Hashimoto, Motoharu Sonogashira, Masaaki Iiyama
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 partially shared weights and VAE to D-CycleGAN (ablation) | D-CycleGAN · Human pose estimation, Avg. error < 10px | accuracy %: 50.8 -> 57.0 | Improved |
| add feature disentanglement into domain-invariant and specific parts (ablation) | CycleGAN+L_fc · Human pose estimation, Avg. error < 10px | accuracy %: 35.5 -> 50.8 | Improved |
| add VAE (KL loss and resampling) without weight sharing (ablation) | D-CycleGAN · Human pose estimation, Avg. error < 10px | accuracy %: 50.8 -> 18.6 | Hurt |
| add partial weight sharing without VAE (ablation) | D-CycleGAN · Human pose estimation, Avg. error < 10px | accuracy %: 50.8 -> 42.1 | Hurt |
| add VAE on top of partially shared auto-encoders (ablation) | PS-AEs · Human pose estimation, Avg. error < 10px | accuracy %: 42.1 -> 57.0 | Improved |
| use PS-VAEs instead of CyCADA | CyCADA · Human pose estimation, Avg. error < 10px | accuracy %: 41.0 -> 57.0 | Improved |
| use PS-VAEs instead of MCD | MCD · Human pose estimation, Avg. error < 10px | accuracy %: 5.3 -> 57.0 | Improved |
| use PS-VAEs instead of CyCADA at heaviest imbalance | CyCADA · MNIST->USPS 50% imbalance | accuracy: 87.6 -> 92.6 | Improved |
| use PS-VAEs instead of UFDN at heaviest imbalance | UFDN · USPS->MNIST 50% imbalance | accuracy: 69.1 -> 82.4 | Improved |
| increase class-1 imbalance from 10% to 50% for PS-VAEs | PS-VAEs · MNIST->USPS | accuracy: 93.9 -> 92.6 | No clear effect |
| increase class-1 imbalance from 10% to 50% for UFDN | UFDN · SVHN->MNIST | accuracy: 91.1 -> 43.6 | Hurt |
| use PS-VAEs instead of MCD on SVHN->MNIST at 10% | MCD · SVHN->MNIST 10% | accuracy: 90.3 -> 73.7 | Hurt |
Closest papers by meaning
- Adversarial Unsupervised Domain Adaptation With Conditional and Label Shift: Infer, Align and Iterate (ICCV 2021 · similarity 0.91)
- Can We Evaluate Domain Adaptation Models Without Target-Domain Labels? (ICLR 2024 · similarity 0.91)
- Inferring Latent Domains for Unsupervised Deep Domain Adaptation (TPAMI 2019 · similarity 0.90)
- Co-regularized Alignment for Unsupervised Domain Adaptation (NeurIPS 2018 · similarity 0.90)
- CASUAL: Conditional Support Alignment for Domain Adaptation with Label Shift (AAAI 2025 · similarity 0.90)
- Prompt-Based Distribution Alignment for Unsupervised Domain Adaptation (AAAI 2024 · similarity 0.89)
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.
