Research map / Graph learning / GNN architectures
Optimization-Induced Graph Implicit Nonlinear Diffusion
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 |
|---|---|---|---|
| replace linear isotropic diffusion with nonlinear (tanh) diffusion (ablation) | GIND · Cornell, Wisconsin, CiteSeer | mean accuracy (%): Cornell +2.34, Wisconsin +3.73, CiteSeer +0.19 | Helped |
| add Laplacian regularization R_Lap to the convex objective (ablation) | GIND · Cora, CiteSeer, PubMed | mean accuracy (%): Cora +0.08, CiteSeer +0.14, PubMed +0.20 | Helped |
| add feature decorrelation regularization R_Dec (ablation) | GIND · Cora, CiteSeer, PubMed | mean accuracy (%): Cora +0.04, CiteSeer +0.03, PubMed +0.06 | Helped |
| use GIND (nonlinear implicit diffusion) instead of EIGNN | node classification on heterophilic graphs · Squirrel | mean accuracy (%): 46.37 -> 56.71 | Helped |
| use phantom gradient backward and variance normalization instead of exact-gradient implicit training | GIND training · PubMed, 500 epochs, 1-layer | training time (s): 47.31 s vs EIGNN 1473.59 s total | Helped |
| GIND does not beat best explicit model GCNII on citation datasets | GIND · Cora, PubMed, PPI | accuracy / micro-F1: Cora 88.25 vs GCNII 88.49; PPI 98.4 vs GCNII 99.5 | Hurt |
Closest papers by meaning
- Advancing Graph Generation through Beta Diffusion (ICLR 2025 · similarity 0.87)
- Learning to Approximate Adaptive Kernel Convolution on Graphs (AAAI 2024 · similarity 0.87)
- Diffusing to the Top: Boost Graph Neural Networks with Minimal Hyperparameter Tuning (ICLR 2025 · similarity 0.87)
- A Generalized Neural Diffusion Framework on Graphs (AAAI 2024 · similarity 0.87)
- Unifying Generation and Prediction on Graphs with Latent Graph Diffusion (NeurIPS 2024 · similarity 0.86)
- Learning Divergence Fields for Shift-Robust Graph Representations (ICML 2024 · similarity 0.86)
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.