Research map / Graph learning / Knowledge and temporal graphs
Fast Mixture of Curvature-Aware Experts for Diverse and Dynamic Graph Topologies
Jiayi Yang, Xing Wei, Chunchun Chen, Yi Feng, Wengang Guo, Rui Fan, Xiaofeng Cao, Xin Sun, Wei Ye
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 |
|---|---|---|---|
| replace GCRouter and ranking constraint with a standard Top-k router (ablation) | DyGMoCE · MOOC, UN Trade, UN Vote | AP | Hurt |
| add ranking constraint to GCRouter (ablation) | DyGMoCE · UN Vote | AP | Improved |
| add geometric encoding (Gromov hyperbolicity, growth rate) to input (ablation) | DyGMoCE · MOOC, UN Trade, UN Vote | AP | Improved |
| use mixed H/E/S expert curvatures instead of a single geometry (ablation) | DyGMoCE · MOOC, UN Trade, UN Vote | AP | Improved |
| increase curvature diversity (Mix-1 to Mix-3) (ablation) | DyGMoCE · MOOC, UN Trade, UN Vote | AP | Improved |
| use fast Riemannian attention instead of conventional Riemannian attention (ablation) | DyGMoCE · MOOC, L=32 | training time: 12 min 28 s -> 9 min 02 s | Improved |
| use fast Riemannian attention instead of conventional Riemannian attention (ablation) | DyGMoCE · MOOC, L=32 | GPU memory: 20,253 MB -> 9,530 MB | Improved |
| use fast Riemannian attention to enable input length 64 (ablation) | DyGMoCE · MOOC and UN Vote, L=64 | memory feasibility: OOM -> 18,627 MB (MOOC) | Improved |
| use Riemannian attention instead of Euclidean attention (ablation) | DyGMoCE · MOOC, L=32 | training time: 6 min 12 s -> 9 min 02 s (EA vs Fast RA) | Hurt |
| vary margin tau and trade-off lambda of ranking loss (ablation) | DyGMoCE · MOOC, UN Trade, UN Vote | AP | No clear effect |
Closest papers by meaning
- Adaptive Riemannian Graph Neural Networks (AAAI 2026 · similarity 0.88)
- GraphMoRE: Mitigating Topological Heterogeneity via Mixture of Riemannian Experts (AAAI 2025 · similarity 0.87)
- Pseudo-Riemannian Graph Transformer (NeurIPS 2025 · similarity 0.87)
- Toward a Unified Geometry Understanding : Riemannian Diffusion Framework for Graph Generation and Prediction (NeurIPS 2025 · similarity 0.86)
- Unifying and Enhancing Graph Transformers via a Hierarchical Mask Framework (NeurIPS 2025 · similarity 0.86)
- Linear Transformer Topological Masking with Graph Random Features (ICLR 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.
