Research map / Label-efficient and robust learning / Domain adaptation and generalization
Adversarial Bayesian Augmentation for Single-Source Domain Generalization
Sheng Cheng, Tejas Gokhale, Yezhou Yang
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 |
|---|---|---|---|
| replace ALT adversarial standard conv with adversarial Bayesian conv (1-layer) | ABA · Digits target avg | accuracy (%): 70.23 -> 74.57 | Improved |
| replace ALT with ABA at 5 layers on PACS | ABA · PACS avg | accuracy (%): 63.60 -> 66.02 | Improved |
| add RandConv module to 1-layer ABA (ablation) | ABA 1-layer · Digits target avg | accuracy (%): 74.57 -> 74.51 | No clear effect |
| add RandConv or Augmix module to ABA (ablation) | ABA 5-layer · PACS avg | accuracy (%): 66.02 -> 66.00 (RandConv), 65.55 (Augmix) | No clear effect |
| add Augmix module to 5-layer ABA (ablation) | ABA 5-layer · Camelyon17 target avg | accuracy (%): 88.19 -> 89.88 | Improved |
| increase ABA from 1 layer to 3 layers (ablation) | ABA · Digits target avg | accuracy (%): 74.57 -> 76.56 | Improved |
| increase ABA from 3 layers to 5 layers (ablation) | ABA · Digits target avg | accuracy (%): 76.56 -> 74.76 | Hurt |
| adversarial learning rate 5e-6 versus 5e-5 or 5e-7 (ablation) | ABA · Digits | accuracy | Hurt |
| increase adversarial steps beyond 5 (ablation) | ABA · Digits | accuracy | No clear effect |
| increase number of Bayesian samplings beyond 10 (ablation) | 1-layer ABA · Digits | accuracy | Hurt |
| apply AdvBNN instead of ERM | baseline · Digits target avg | accuracy (%): 53.50 -> 51.92 | Hurt |
| apply RandConv instead of ERM on subpopulation shift | baseline · Living17 target | accuracy (%): 70.97 -> 61.18 | Hurt |
Closest papers by meaning
- Out-of-Domain Generalization From a Single Source: An Uncertainty Quantification Approach (TPAMI 2022 · similarity 0.90)
- Learning to Learn Single Domain Generalization (CVPR 2020 · similarity 0.89)
- A Simple Feature Augmentation for Domain Generalization (ICCV 2021 · similarity 0.89)
- AdvST: Revisiting Data Augmentations for Single Domain Generalization (AAAI 2024 · similarity 0.89)
- Generalizing to Unseen Domains via Adversarial Data Augmentation (NeurIPS 2018 · similarity 0.89)
- Adversarial Domain Prompt Tuning and Generation for Single Domain Generalization (CVPR 2025 · similarity 0.86)
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.
