Research map / Probabilistic and scientific modeling / Variational and generative sampling
Partial Distribution Matching via Partial Wasserstein Adversarial Networks
Zi-Ming Wang, Nan Xue, Ling Lei, Rebecka J. Jörnsten, Gui-Song Xia
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 four extra techniques (label smoothing, complement objective, NRC, alpha-power) (ablation) | PWAN · OfficeHome partial DA (avg) | accuracy (%): 77.2 -> 80.5 | Improved |
| add four extra techniques (ablation) | PWAN · DomainNet partial DA (avg) | accuracy (%): 66.8 -> 74.7 | Improved |
| align only a fraction of reference features (m_alpha=10 instead of complete alignment m_alpha=1) (ablation) | PWAN · AR-OfficeHome partial DA | negative transfer | Improved |
| add entropy regularizer (lambda_2 > 0) (ablation) | PWAN · AR-OfficeHome partial DA | feature alignment | Improved |
| add PW alignment (lambda_1 > 0) versus reference-only training (ablation) | PWAN · AR-OfficeHome partial DA | feature alignment | Improved |
| increase network update frequency u (ablation) | PWAN · AC-OfficeHome partial DA | accuracy and training time: accuracy saturates beyond u=10; training time up 10% from u=1 to u=20 | Mixed |
| increase network update frequency u (ablation) | PWAN registration · bunny point set registration (2000 points) | MSE | Improved |
| vary m_alpha tilde over [10^0, 10^2] (ablation) | PWAN · ImageNet-Caltech partial DA | accuracy: variation about 2% | No clear effect |
| use PW divergence instead of KL or L2 under outliers | PW divergence · toy 1D registration | outlier robustness | Improved |
| use m-PWAN instead of BCPD | registration · UWA object rigid registration | median rotation error (deg): 0.10 -> 0.10 | No clear effect |
| use m-PWAN instead of BCPD | registration · ASL outdoor rigid registration | median rotation error (deg): 8.08 -> 2.00 | Improved |
| use PWAN instead of ADV | partial DA · VisDa17 | accuracy (%): 80.1 -> 84.8 | Improved |
Closest papers by meaning
- Partial Wasserstein Adversarial Network for Non-rigid Point Set Registration (ICLR 2022 · similarity 0.91)
- A Unified Wasserstein Distributional Robustness Framework for Adversarial Training (ICLR 2022 · similarity 0.86)
- Iterative Feature Matching: Toward Provable Domain Generalization with Logarithmic Environments (NeurIPS 2022 · similarity 0.86)
- Multi-marginal Wasserstein GAN (NeurIPS 2019 · similarity 0.85)
- ALICE: Towards Understanding Adversarial Learning for Joint Distribution Matching (NeurIPS 2017 · similarity 0.85)
- Expressive Score-Based Priors for Distribution Matching with Geometry-Preserving Regularization (ICML 2025 · similarity 0.85)
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.
