Research map / Label-efficient and robust learning / Domain adaptation and generalization
DNA: Domain Generalization with Diversified Neural Averaging
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 |
|---|---|---|---|
| add SWA weight averaging to PJS-loss training (ablation) | PJS loss model · OfficeHome, TerraIncognita, DomainNet | out-of-domain accuracy: +4.0 OfficeHome, +2.0 TerraIncognita, +2.8 DomainNet | Helped |
| add diversity term DP (explicit diversity enforcement) (ablation) | DNA · PACS, VLCS, DomainNet | out-of-domain accuracy: -0.2 PACS, -0.3 VLCS, -0.4 DomainNet | Hurt |
| add diversity term DP (explicit diversity enforcement) (ablation) | DNA Bernoulli · OfficeHome, TerraIncognita | out-of-domain accuracy: +0.4 OfficeHome, +0.4 TerraIncognita | Helped |
| use Bernoulli instead of Gaussian dropout variational family (ablation) | DNA with DP · OfficeHome, TerraIncognita | out-of-domain accuracy: +0.3 OfficeHome, +0.3 TerraIncognita | Helped |
| replace PJS loss with cross-entropy (SWA and DP kept) (ablation) | DNA · OfficeHome, TerraIncognita, DomainNet | out-of-domain accuracy: -0.1 OfficeHome, -0.3 TerraIncognita, -0.3 DomainNet | Hurt |
| classifier averaging over 20 saved checkpoints instead of single-run weight averaging (ablation) | DNA · OfficeHome, TerraIncognita | out-of-domain accuracy: -0.5 OfficeHome, -1.2 TerraIncognita (K=20 vs DNA) | Hurt |
| DNA (SWA-based fast ensemble) vs SWAD and ERM average over five benchmarks | DNA · five DG benchmarks average | average accuracy: +0.7 over SWAD, +4.3 over ERM | Helped |
| full DNA method vs baselines on VLCS | DNA · VLCS | accuracy: 79.0 vs 79.1 SWAD | No clear effect |
Closest papers by meaning
- Understanding Domain Generalization: A Noise Robustness Perspective (ICLR 2024 · similarity 0.87)
- Set Valued Predictions For Robust Domain Generalization (ICML 2025 · similarity 0.85)
- Improving Domain Generalization with Domain Relations (ICLR 2024 · similarity 0.85)
- Rethinking Multi-domain Generalization with A General Learning Objective (CVPR 2024 · similarity 0.85)
- One-Step Generalization Ratio Guided Optimization for Domain Generalization (ICML 2025 · similarity 0.85)
- Lost Domain Generalization Is a Natural Consequence of Lack of Training Domains (AAAI 2024 · similarity 0.84)
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.