Research map / Graph learning / GNN architectures
GMNN: Graph Markov Neural Networks
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 |
|---|---|---|---|
| add a CRF label-dependency model trained alongside GNN inference (ablation) | GMNN · Cora, Citeseer, Pubmed unsupervised node representation learning | accuracy: 78.1 -> 82.8; 68.0 -> 71.5; 79.3 -> 81.6 | Helped |
| add object attributes to pφ (ablation) | GMNN · Cora, Citeseer, Pubmed object classification | accuracy: 83.4 -> 83.7; 73.1 -> 72.9; 81.4 -> 81.8 | Mixed |
| use amortized GNN posterior inference instead of independent categorical parameters (ablation) | GMNN · Cora, Citeseer, Pubmed | accuracy: 45.3 -> 83.4; 28.1 -> 73.1; 42.2 -> 81.4 | Helped |
| use two GC layers rather than one in inference (ablation) | GMNN · Cora, Citeseer, Pubmed | accuracy: 72.9 -> 83.4; 67.6 -> 73.1; 71.8 -> 81.4 | Helped |
| increase inference depth from two to three GC layers (ablation) | GMNN · Cora, Citeseer, Pubmed | accuracy: 83.4 -> 82.0; 73.1 -> 70.6; 81.4 -> 80.7 | Hurt |
| use a learned GNN learning network instead of fixed mean-pooling label propagation (ablation) | GMNN · Cora, Citeseer, Pubmed object classification | accuracy: 82.4 -> 83.4; 71.9 -> 73.1; 80.7 -> 81.4 | Helped |
| use two rather than one GC layer in the learning network (ablation) | GMNN · Cora, Citeseer, Pubmed object classification | accuracy: 83.1 -> 83.4; 73.1 -> 73.1; 80.9 -> 81.4 | Mixed |
| use three rather than two GC layers in the learning network (ablation) | GMNN · Cora, Citeseer, Pubmed object classification | accuracy: 83.4 -> 83.6; 73.1 -> 73.0; 81.4 -> 81.5 | Mixed |
| use annealing for estimating the mean-field expectation instead of single sampling (ablation) | GMNN · Cora, Citeseer, Pubmed object classification | accuracy: 82.1 -> 83.4; 71.5 -> 73.1; 80.4 -> 81.4 | Helped |
| use multiple samples rather than one sample for expectation estimation (ablation) | GMNN · Cora, Citeseer, Pubmed object classification | accuracy: 82.1 -> 83.2; 71.5 -> 72.5; 80.4 -> 81.1 | Helped |
| use max pooling rather than single sampling for expectation estimation (ablation) | GMNN · Cora, Citeseer, Pubmed object classification | accuracy: 82.1 -> 83.2; 71.5 -> 72.8; 80.4 -> 81.2 | Helped |
| use mean pooling rather than single sampling for expectation estimation (ablation) | GMNN · Cora, Citeseer, Pubmed object classification | accuracy: 82.1 -> 83.4; 71.5 -> 72.6; 80.4 -> 80.5 | Mixed |
Closest papers by meaning
- A Flexible Generative Framework for Graph-based Semi-supervised Learning (NeurIPS 2019 · similarity 0.88)
- Graph Stochastic Neural Networks for Semi-supervised Learning (NeurIPS 2020 · similarity 0.88)
- Bayesian Graph Convolutional Neural Networks for Semi-Supervised Classification (AAAI 2019 · similarity 0.88)
- Semi-Supervised Classification with Graph Convolutional Networks (ICLR 2017 · similarity 0.88)
- Predicting Global Label Relationship Matrix for Graph Neural Networks under Heterophily (NeurIPS 2023 · similarity 0.87)
- Hypergraph-enhanced Dual Semi-supervised Graph Classification (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.