Research map / Label-efficient and robust learning / Domain adaptation and generalization
Domain Adaptation with Asymmetrically-Relaxed Distribution Alignment
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 |
|---|---|---|---|
| replace exact latent-distribution matching with asymmetrically relaxed alignment | DANN · Synthetic target (label shift) | accuracy: 59.1$\pm$5.1 -> 99.9$\pm$ 0.1 | Helped |
| replace exact latent-distribution matching with asymmetrically relaxed alignment | DANN · MNIST→USPS target [0-4] (shift) | accuracy: 50.0$\pm$1.9 -> 81.0$\pm$1.6 | Helped |
| replace exact latent-distribution matching with asymmetrically relaxed alignment | DANN · USPS→MNIST target [5-9] (shift) | accuracy: 37.1$\pm$3.5 -> 45.4$\pm$7.1; 37.1$\pm$3.5 -> 34.7$\pm$7.1 | Mixed |
| replace exact latent-distribution matching with asymmetrically relaxed alignment | DANN · MNIST→USPS target [0-9] (no shift) | accuracy: 78.5$\pm$1.6 -> 82.0$\pm$0.4; 78.5$\pm$1.6 -> 72.3$\pm$1.2 | Mixed |
| replace exact latent-distribution matching with asymmetrically relaxed alignment | DANN · USPS→MNIST target [0-9] (no shift) | accuracy: 81.9$\pm$6.7 -> 58.5$\pm$2.2 | Hurt |
| vary the relaxed-alignment threshold β (ablation) | β-admissible DANN variants · Synthetic target (label shift) | accuracy: fDANN-$\beta$: 66.0$\pm$ 41.6 -> 99.9$\pm$ 0.0; WDANN1-$\beta$: 45.7$\pm$ 41.5 -> 99.9$\pm$0.0 | Mixed |
| vary the relaxed-alignment threshold β (ablation) | fDANN · USPS→MNIST target [0-9] (no shift) | accuracy: 75.4$\pm$4.5; 70.0$\pm$3.3; 58.5$\pm$2.2 | Hurt |
| use implicit reweighting by sorting for relaxed JS divergence | DANN · MNIST→USPS target [0-9] (no shift) | accuracy: 78.5$\pm$1.6 -> 82.0$\pm$0.4 | Helped |
| use implicit reweighting by sorting for relaxed JS divergence | DANN · USPS→MNIST target [5-9] (shift) | accuracy: 37.1$\pm$3.5 -> 45.4$\pm$7.1; 37.1$\pm$3.5 -> 36.1$\pm$5.2 | Mixed |
| use relaxed Wasserstein distance with soft-plus critic output | WDANN · Synthetic target (label shift) | accuracy: 50.8$\pm$32.1 -> 99.9$\pm$0.0; 50.8$\pm$32.1 -> 45.7$\pm$ 41.5 | Mixed |
| use relaxed Wasserstein distance with RELU critic output | WDANN · Synthetic target (label shift) | accuracy: 50.8$\pm$32.1 -> 99.7$\pm$ 0.2 | Helped |
| implicit reweighting by sorting rather than explicit weight parameterization (ablation) | Relaxed distribution distance · Synthetic target (label shift) | accuracy: sDANN-$\beta$: 99.9$\pm$ 0.1; sWDANN-$\beta$: 79.0$\pm$ 5.9 | Mixed |
Closest papers by meaning
- Domain Adaptation with Conditional Distribution Matching and Generalized Label Shift (NeurIPS 2020 · similarity 0.91)
- KL Guided Domain Adaptation (ICLR 2022 · similarity 0.90)
- CASUAL: Conditional Support Alignment for Domain Adaptation with Label Shift (AAAI 2025 · similarity 0.90)
- Implicit Class-Conditioned Domain Alignment for Unsupervised Domain Adaptation (ICML 2020 · similarity 0.90)
- Adversarial Support Alignment (ICLR 2022 · similarity 0.90)
- Co-regularized Alignment for Unsupervised Domain Adaptation (NeurIPS 2018 · similarity 0.90)
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.