Research map / Graph learning / GNN architectures
pathGCN: Learning General Graph Spatial Operators from Paths
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 |
|---|---|---|---|
| Learn a flexible spatial operator from paths rather than use a fixed Laplacian-based operator | GCNs · Cora, Citeseer, and Pubmed (semi-supervised node classification) | accuracy (%): 81.0 -> 82.7 | Helped |
| Learn a flexible spatial operator from paths rather than use a fixed or proxy spatial operator | GCNs · Fully-supervised node classification benchmarks | accuracy (%): 88.49 -> 90.02 | Helped |
| Use path-based learned spatial kernels | GCNs · PPI (inductive learning) | micro-averaged F1: 99.58 -> 99.61 | Helped |
| Use path-based learned spatial kernels | Graph classification models · MUTAG, PTC, and PROTEINS | accuracy (%): 92.7_{\pm6.1} -> 94.7_{\pm4.7}; 68.2_{\pm5.6} -> 75.2_{\pm5.3}; 77.0_{\pm4.3} -> 80.4_{\pm4.2} | Helped |
| Use path-based learned spatial kernels | Graph classification models · NCI1 | accuracy (%): 83.6_{\pm1.4} -> 83.3_{\pm1.3} | Hurt |
| Use a global learned spatial operator shared across layers and channels instead of depth-wise operators (ablation) | pathGCN · Cora, Citeseer, and Pubmed (semi-supervised node classification) | accuracy (%): Cora: 84.2 -> 83.0; Citeseer: 74.3 -> 73.1; Pubmed: 81.8 -> 80.9 | Hurt |
| Use a per-layer spatial operator shared across channels instead of depth-wise operators (ablation) | pathGCN · Cora, Citeseer, and Pubmed (semi-supervised node classification) | accuracy (%): Cora: 84.2 -> 82.9; Citeseer: 74.3 -> 73.4; Pubmed: 81.8 -> 81.1 | Hurt |
| Increase the number of sampled paths beyond 20 (ablation) | pathGCN · Cora, Citeseer, and Pubmed (semi-supervised node classification) | accuracy (%) | Hurt |
| Increase path length from 1 toward 7 (ablation) | pathGCN · Cora, Citeseer, and Pubmed (semi-supervised node classification) | accuracy (%) | Helped |
| Increase path length beyond the useful range (ablation) | pathGCN · Cora, Citeseer, and Pubmed (semi-supervised node classification) | accuracy (%) | Mixed |
| Use stochastic rather than deterministic pathGCN inference (ablation) | pathGCN · Cora, Citeseer, Pubmed, ogbn-arxiv, Wisconsin, and PPI | accuracy (%) or micro-averaged F1: 85.8 vs 85.8; 75.8 vs 75.8; 82.7 vs 82.7; 72.84 vs 72.83; 93.51 vs 93.53; 99.61 vs 99.61 | No clear effect |
Closest papers by meaning
- Deeper Insights Into Graph Convolutional Networks for Semi-Supervised Learning (AAAI 2018 · similarity 0.90)
- L2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks (CVPR 2020 · similarity 0.90)
- Path Integral Based Convolution and Pooling for Graph Neural Networks (NeurIPS 2020 · similarity 0.89)
- CKGConv: General Graph Convolution with Continuous Kernels (ICML 2024 · similarity 0.89)
- Simplifying Graph Convolutional Networks (ICML 2019 · similarity 0.89)
- Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · similarity 0.89)
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.