Research map / Label-efficient and robust learning / Domain adaptation and generalization
Gradient Distribution Alignment Certificates Better Adversarial Domain Adaptation
Zhiqiang Gao, Shufei Zhang, Kaizhu Huang, Qiufeng Wang, Chaoliang Zhong
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 feature gradient alignment on top of DANN (ablation) | DANN · Office-31 (average) | accuracy: 82.2 -> 88.9 | Improved |
| add feature gradient alignment on top of CDAN-E (ablation) | CDAN-E · Office-31 (average) | accuracy: 87.7 -> 89.2 | Improved |
| add feature gradient alignment on top of MDD (ablation) | MDD · Office-31 (average) | accuracy: 88.9 -> 90.6 | Improved |
| add feature gradient alignment on top of MDD (ablation) | MDD · Office-Home (average) | accuracy: 68.1 -> 71.5 | Improved |
| add self-supervised pseudo-labeling (ablation) | FGDA · Office-31 (average) | accuracy: 86.0 -> 88.6 | Improved |
| add Jacobian regularization (ablation) | FGDA · Office-31 (average) | accuracy: 88.3 -> 88.6 | Improved |
| add both Jacobian regularization and pseudo-labeling (ablation) | FGDA · Office-31 (average) | accuracy: 85.0 -> 88.6 | Improved |
| gradient alignment alone replaces DANN feature alignment (ablation) | FGDA w/o FJR, SPL · Office-31 (average) | accuracy: 82.2 -> 85.0 | Improved |
| use cross-entropy instead of conditional entropy for target gradients (ablation) | FGDA+DANN · Office-31 (average) | accuracy: 87.6 -> 88.9 | Improved |
| use FGDA alone instead of MDD | FGDA · Office-31 (average) | accuracy: 88.9 -> 88.6 | No clear effect |
| use FGDA alone instead of MDD | FGDA · Office-Home (average) | accuracy: 68.1 -> 68.3 | No clear effect |
| vary Jacobian regularization weight lambda2 from 0.005 to 0.010 (ablation) | FGDA w/o SPL · Office-31 A->W | accuracy: 85.0 -> 91.3 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DCAN — DCAN: Dual Channel-wise Alignment Networks for Unsupervised Scene Adaptation (ECCV 2018) · better on 2 of 2 shared results
- DADA — DADA: Depth-Aware Domain Adaptation in Semantic Segmentation (ICCV 2019) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Domain Adaptation with Asymmetrically-Relaxed Distribution Alignment (ICML 2019 · similarity 0.91)
- f-Domain Adversarial Learning: Theory and Algorithms (ICML 2021 · similarity 0.89)
- Unsupervised Domain Adaptation via Regularized Conditional Alignment (ICCV 2019 · similarity 0.88)
- Cross-Domain Gradient Discrepancy Minimization for Unsupervised Domain Adaptation (CVPR 2021 · similarity 0.88)
- Multi-Adversarial Domain Adaptation (AAAI 2018 · similarity 0.88)
- CASUAL: Conditional Support Alignment for Domain Adaptation with Label Shift (AAAI 2025 · similarity 0.87)
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.
