Research map / Label-efficient and robust learning / Domain adaptation and generalization
Shape-Biased Domain Generalization via Shock Graph Embeddings
Maruthi Narayanan, Vickram Rajendran, Benjamin Kimia
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 TAG convolution instead of GAT (ablation) | shock graph GNN · STL-10 | accuracy: 56.2% -> 58% | Improved |
| use TAG convolution instead of GCN (ablation) | shock graph GNN · STL-10 | accuracy: 54.3% -> 58% | Improved |
| replace ResNet-50 image CNN (DeepALL) with shock graph GNN | DeepALL · PACS average | accuracy: 35.80 -> 47.45 | Improved |
| replace ResNet-50 image CNN (DeepALL) with shock graph GNN | DeepALL · VLCS average | accuracy: 50.60 -> 57.16 | Improved |
| replace ResNet-50 image CNN (DeepALL) with shock graph GNN | DeepALL · PACS Sketch | accuracy: 24.00 -> 54.15 | Improved |
| replace image CNN DG methods with shock graph GNN on color-shifted domain | DeepALL · Colored MNIST rho=0.9 | accuracy: 10.3 -> 71.60 | Improved |
| replace image CNN with shock graph GNN on texture-confounded Art domain | Self Challenge · PACS Art | accuracy: 42.38 -> 33.26 | Hurt |
| replace image CNN with shock graph GNN on Photo domain | Self Challenge · PACS Photo | accuracy: 55.02 -> 53.23 | Hurt |
| use ImageNet-pretrained weights for image CNN methods | image CNN DG methods · Colored MNIST | accuracy | No clear effect |
| replace MLDG with shock graph GNN | MLDG · VLCS average | accuracy: 53.50 -> 57.16 | Improved |
| replace Self Challenge with shock graph GNN | Self Challenge · PACS average | accuracy: 46.95 -> 47.45 | Improved |
| replace MLDG with shock graph GNN | MLDG · Colored MNIST average | accuracy: 51.6 -> 71.94 | Improved |
Closest papers by meaning
- Shape-Texture Debiased Neural Network Training (ICLR 2021 · similarity 0.87)
- Shape or Texture: Understanding Discriminative Features in CNNs (ICLR 2021 · similarity 0.86)
- ImageNet-trained CNNs are not biased towards texture: Revisiting feature reliance through controlled suppression (NeurIPS 2025 · similarity 0.85)
- Can Biases in ImageNet Models Explain Generalization? (CVPR 2024 · similarity 0.84)
- The Shape of Events: Edge-Based Inductive Biases via Cross-Domain Distillation (NeurIPS 2026 · similarity 0.84)
- The Effect of Intrinsic Dataset Properties on Generalization: Unraveling Learning Differences Between Natural and Medical Images (ICLR 2024 · similarity 0.83)
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.
