Research map / Label-efficient and robust learning / Domain adaptation and generalization
When Neural Networks Fail to Generalize? A Model Sensitivity Perspective
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 |
|---|---|---|---|
| Use 3SADA+0Mix instead of L2D | ConvNet · DIGITS (MNIST source; target-domain average) | accuracy: 74.45 -> 76.56 | Helped |
| Use 2SADA+1Mix instead of L2D | ResNet-18 · PACS (single-source; target-domain average) | accuracy: 64.74 -> 66.18 | Helped |
| Use 2SADA+1Mix instead of AugMix | WRN-16-4 · CIFAR-10-C (severity 5; corruption-category average) | accuracy: 75.28 -> 77.83 | Helped |
| Add two SADA-augmented images (ablation) | 2SADA+1Mix on ConvNet · DIGITS (MNIST source; target-domain average) | accuracy: 60.27 -> 76.38 | Helped |
| Add AugMix (ablation) | 2SADA+1Mix on ConvNet · DIGITS (MNIST source; target-domain average) | accuracy: 69.46 -> 76.38 | Helped |
| Add the JS divergence consistency loss (ablation) | 2SADA+1Mix on ConvNet · DIGITS (MNIST source; target-domain average) | accuracy: 70.37 -> 76.38 | Helped |
| Replace the proposed enhanced sensitivity map with the original sensitivity map (ablation) | 3SADA+0Mix · PACS and DIGITS single-domain generalization | accuracy | Hurt |
| Replace targeted adversarial spectral augmentation with random spectral perturbation (ablation) | 3SADA+0Mix · PACS and DIGITS single-domain generalization | accuracy | Hurt |
| Vary the JS-loss weighting factor within [0.1, 0.6] (ablation) | 2SADA+1Mix · PACS | accuracy | No clear effect |
| Increase perturbation strength through perturbation steps and step size (ablation) | 2SADA+1Mix · PACS | accuracy | Mixed |
| Increase random initialization strength beyond ε < 0.30 (ablation) | 2SADA+1Mix · PACS | accuracy | Hurt |
Closest papers by meaning
- Sharpness-Aware Model-Agnostic Long-Tailed Domain Generalization (AAAI 2024 · similarity 0.90)
- Gradient-Guided Annealing for Domain Generalization (CVPR 2025 · similarity 0.88)
- Set Valued Predictions For Robust Domain Generalization (ICML 2025 · similarity 0.86)
- AdvST: Revisiting Data Augmentations for Single Domain Generalization (AAAI 2024 · similarity 0.86)
- PracticalDG: Perturbation Distillation on Vision-Language Models for Hybrid Domain Generalization (CVPR 2024 · similarity 0.86)
- 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.