Research map / Label-efficient and robust learning / Domain adaptation and generalization
Understanding Hessian Alignment for Domain Generalization
Sobhan Hemati, Guojun Zhang, Amir Estiri, Xi Chen
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 |
|---|---|---|---|
| use Hutchinson Hessian-diagonal alignment instead of Fishr/V-REx | Colored MNIST MLP · Colored MNIST | test accuracy (%): 69.5 -> 74.0 | Improved |
| use Hutchinson alignment under heavy label shift | Imbalanced Colored MNIST · Imbalanced Colored MNIST | test accuracy (%): 35.5 -> 47.7 | Improved |
| use HGP (Hessian-gradient product) instead of Fishr | Colored MNIST · Colored MNIST | test accuracy (%): 69.5 -> 69.4 | No clear effect |
| use HGP on DomainBed instead of ERM | ResNet-50 DomainBed · DomainBed average | average accuracy (%): 66.6 -> 66.9 | No clear effect |
| use Hutchinson alignment on DomainBed instead of ERM | ResNet-50 DomainBed · DomainBed average | average accuracy (%): 66.6 -> 68.5 | Improved |
| add Hessian alignment on top of gradient alignment (ablation) | Hutchinson · PACS (test domain sketch) | test accuracy (%): 77.0 -> 81.4 | Improved |
| add gradient alignment on top of Hessian alignment (ablation) | Hutchinson · PACS (test domain sketch) | test accuracy (%): 79.4 -> 81.4 | Improved |
| add gradient alignment term (beta) to Hutchinson (ablation) | Hutchinson · Colored MNIST | test accuracy (%): 9.8 -> 74.0 | Improved |
| add Hessian alignment term (alpha) to Hutchinson (ablation) | Hutchinson · Colored MNIST | test accuracy (%): 69.3 -> 74.0 | Improved |
| add Hessian-gradient-product term (alpha) to HGP (ablation) | HGP · Colored MNIST | test accuracy (%): 69.3 -> 69.4 | No clear effect |
| use Hutchinson Hessian alignment instead of baselines | CMNIST training cost · CMNIST training time, 100 steps | training time: ERM 15.3 -> Hutchinson 62.4 | Hurt |
Closest papers by meaning
- Investigating the Overlooked Hessian Structure: From CNNs to LLMs (ICML 2025 · similarity 0.86)
- Towards Robust Out-of-Distribution Generalization Bounds via Sharpness (ICLR 2024 · similarity 0.86)
- HYPO: Hyperspherical Out-Of-Distribution Generalization (ICLR 2024 · similarity 0.85)
- Neglected Hessian component explains mysteries in sharpness regularization (NeurIPS 2024 · similarity 0.85)
- OOD-Chameleon: Is Algorithm Selection for OOD Generalization Learnable? (ICML 2025 · similarity 0.85)
- Gradient-Weight Alignment as a Train-Time Proxy for Generalization in Classification Tasks (NeurIPS 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.
