Research map / Graph learning / GNN architectures
FlowSymm: Physics–Aware, Symmetry–Preserving Graph Attention for Network Flow Completion
Ege Demirci, Francesco Bullo, Ananthram Swami, Ambuj Singh
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 |
|---|---|---|---|
| add edge-wise attention over group-action basis instead of uniform random actions with one global weight (ablation) | FlowSymm (k=256) · Traffic, Power, Bike | RMSE: +5% (Traffic), +11.5% (Power), +6.8% (Bike) | Improved |
| add group-action basis modeling to attention encoder (ablation) | Bil-GAT · Traffic | RMSE: 0.062 -> 0.057 | Improved |
| add group-action basis modeling to attention encoder (ablation) | Bil-GAT · Power | RMSE: 0.029 -> 0.026 | Improved |
| add implicit bilevel optimization (ablation) | GAT · Power | RMSE: 0.065 -> 0.026 | Improved |
| add implicit bilevel optimization (ablation) | GAT · Traffic | RMSE: 0.067 -> 0.057 | Improved |
| increase basis size from 64 to 256 (ablation) | FlowSymm · Traffic | RMSE: 0.060 -> 0.057 | Improved |
| increase basis size from 256 to 512 (ablation) | FlowSymm · Traffic, Power, Bike | RMSE: 0.057 -> 0.057 | No clear effect |
| learn coefficients over valid basis instead of predicting then projecting (ablation) | FlowSymm · Traffic | RMSE: 0.201 -> 0.057 | Improved |
| replace Bil-GCN diagonal regularizer with FlowSymm group actions | Bil-GCN · Traffic | RMSE: 0.062 -> 0.057 | Improved |
| replace Bil-GCN diagonal regularizer with FlowSymm group actions | Bil-GCN · Traffic | MAE: 0.034 -> 0.028 | Improved |
| add 50% Gaussian noise to injection vector c at test time | FlowSymm · Traffic | RMSE: 0.057 -> 0.060 | Hurt |
| add 50% Gaussian noise to injection vector c at test time | FlowSymm · Power | RMSE: 0.026 -> 0.028 | Hurt |
Closest papers by meaning
- Topology-aware Neural Flux Prediction Guided by Physics (ICML 2025 · similarity 0.85)
- Flowette: Flow Matching with Graphette Priors for Graph Generation (NeurIPS 2026 · similarity 0.85)
- Structured Coupling for Flow Matching (NeurIPS 2026 · similarity 0.85)
- Symmetry-Aware GFlowNets (ICML 2025 · similarity 0.85)
- Geometry-Aware Edge Pooling for Graph Neural Networks (NeurIPS 2025 · similarity 0.85)
- FlowFeat: Pixel-Dense Embedding of Motion Profiles (NeurIPS 2025 · 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.
