Research map / Graph learning / Graph algorithms and clustering
Slow Learning and Fast Inference: Efficient Graph Similarity Computation via Knowledge Distillation
Can Qin, Handong Zhao, Lichen Wang, Huan Wang, Yulun Zhang, Yun Fu
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 |
|---|---|---|---|
| distill joint embeddings directly instead of pseudo individual embeddings (ablation) | student GSC model · GED AIDS | MSE (x1e-3): 1.546 -> 2.258 | Hurt |
| add knowledge distillation from teacher (ablation) | student siamese model · GED AIDS | MSE (x1e-3): 1.770 -> 1.546 | Improved |
| add knowledge distillation from teacher (ablation) | student siamese model · GED IMDB | MSE (x1e-3): 0.763 -> 0.581 | Improved |
| use both first and second order distance instead of first order only (ablation) | student KD · GED AIDS | MSE (x1e-3): 1.604 -> 1.546 | Mixed |
| add attention in embedding fusion network (ablation) | teacher · GED AIDS | MSE (x1e-3): 1.762 -> 1.601 | Improved |
| use GIN backbone instead of GCN (ablation) | teacher · GED AIDS | MSE (x1e-3): 2.158 -> 1.601 | Improved |
| use multi-level features instead of single level (ablation) | teacher · GED AIDS | MSE (x1e-3): 1.824 -> 1.601 | Improved |
| add reconstructed joint embedding regression loss L'reg (ablation) | student KD · GED AIDS | MSE (x1e-3): 1.711 -> 1.546 | Improved |
| use student with offline stored embeddings instead of teacher/raw-graph inference | student · AIDS inference time | time: 11.139s -> 0.148s | Improved |
| student vs teacher accuracy | EGSC · GED AIDS | MSE (x1e-3): 1.601 -> 1.546 | Mixed |
Closest papers by meaning
- GraSP: Simple Yet Effective Graph Similarity Predictions (AAAI 2025 · similarity 0.90)
- Graph Classification via Reference Distribution Learning: Theory and Practice (NeurIPS 2024 · similarity 0.89)
- Position: Graph Matching Systems Deserve Better Benchmarks (ICML 2025 · similarity 0.88)
- Exploring Consistency in Graph Representations: from Graph Kernels to Graph Neural Networks (NeurIPS 2024 · similarity 0.88)
- Neural Gaussian Similarity Modeling for Differential Graph Structure Learning (AAAI 2024 · similarity 0.87)
- GC4NC: A Benchmark Framework for Graph Condensation on Node Classification with New Insights (NeurIPS 2025 · similarity 0.87)
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.
