Research map / Label-efficient and robust learning / OOD and anomaly detection
Topology-aware Robust Optimization for Out-of-Distribution Generalization
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 topology-aware prior (betweenness centrality constraint) to group DRO (TRO) vs DRO | group DRO on DG-15/DG-60 · DG-15 and DG-60 synthetic classification | Accuracy (%): DG-15: 43.22 -> 67.89; DG-60: 79.59 -> 90.72 | Helped |
| Use data-driven topology (diffusion EMD) rather than physical adjacency topology (ablation) | TRO · DG-15 / DG-60 / TPT-48 regression | Accuracy (%) / MSE: DG-15 67.56 -> 67.89; DG-60 89.19 -> 90.72; TPT-48 E->W 1.466 -> 1.458; N->S 1.177 -> 1.129 | Helped |
| Topology-constrained robust optimization vs DRO | TRO on TPT-48 temperature regression · N(24)->S(24) all test groups | MSE: 1.356 -> 1.129 | Helped |
| Topology-constrained robust optimization vs DRO | TRO on TPT-48 temperature regression · E(24)->W(24) all test groups | MSE: 1.684 -> 1.458 | Helped |
| Topology-aware robust optimization vs ERM on distant groups | TRO on TPT-48 · N(24)->S(24) Hop-3 groups | MSE: REx 2.266 vs TRO 1.604 | Helped |
| Importance-weighted ERM (fixed topology weights, no worst-case component) (ablation) | ERM with topology weights · N(24)->S(24) TPT-48 | MSE: inferior to ERM and DRO (exact values not in main text) | Hurt |
| Topology-aware robust optimization vs ERM on PACS | TRO on PACS · PACS average accuracy | Accuracy (%): 85.7 -> 86.5 (ERM); 86.0 -> 86.5 vs CORAL SOTA | Helped |
| Topology-aware robust optimization on Terra | TRO on Terra · Terra average accuracy | Accuracy (%): 47.2 -> 49.0 vs ERM; 49.3 -> 49.0 vs MMD (SOTA) | Mixed |
| Topology-aware robust optimization on VLCS | TRO on VLCS · VLCS average accuracy | Accuracy (%): TRO 77.2 equals DRO 77.2; below ERM-level/DANN 78.7 SOTA | No clear effect |
| Topology-aware robust optimization on Sen1Floods11 flood segmentation | TRO on Sen1Floods11 · held-out BOL event IoU | IoU: TRO achieves the highest IoU on the test set (0.637 vs 0.633 DRO, 0.57 IRM, 0.568 ERM per excluded table) | Helped |
| Sensitivity of TRO to dual variable lambda on Sen1Floods11 (ablation) | TRO · Sen1Floods11 | IoU: IoU stable across wide range of lambda; 0.01 best | No clear effect |
Closest papers by meaning
- Topological Generalization Bounds for Discrete-Time Stochastic Optimization Algorithms (NeurIPS 2024 · similarity 0.88)
- Ensemble Pruning for Out-of-distribution Generalization (ICML 2024 · similarity 0.87)
- Towards Robust Out-of-Distribution Generalization Bounds via Sharpness (ICLR 2024 · similarity 0.87)
- Topology-Aware Dynamic Reweighting for Distribution Shifts on Graph (ICML 2025 · similarity 0.87)
- Gradual Domain Adaptation via Manifold-Constrained Distributionally Robust Optimization (NeurIPS 2024 · similarity 0.86)
- Graph Structure Extrapolation for Out-of-Distribution Generalization (ICML 2024 · 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.