Research map / Graph learning / GNN architectures
Semi-Supervised Learning With Graph Learning-Convolutional Networks
Check 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 the graph jointly with graph convolution instead of using a fixed graph | GCN · Citeseer, Cora, Pubmed | accuracy: 70.9% -> 72.0%; 82.9% -> 85.5%; 77.9% -> 78.3% | Helped |
| learn the graph jointly with graph convolution instead of using a fixed graph | GCN · SVHN | accuracy: 71.33±1.48% -> 79.14±0.38%; 73.43±0.46% -> 80.68±0.22%; 73.63±0.52% -> 81.43±0.34% | Helped |
| learn the graph jointly with graph convolution instead of using a fixed graph | GCN · CIFAR10 | accuracy: 60.43±0.56% -> 66.67±0.24%; 60.91±0.50% -> 69.33±0.54%; 60.99±0.49% -> 70.39±0.54% | Helped |
| learn the graph jointly with graph convolution instead of using a fixed graph | GCN · MNIST | accuracy: 90.59±0.26% -> 94.28±0.28%; 90.91±0.19% -> 95.09±0.17%; 91.01±0.23% -> 95.46±0.20% | Helped |
| learn the graph jointly with graph convolution instead of using a fixed graph | GCN · Scene15 | accuracy: 91.42±2.07% -> 96.19±0.38%; 94.41±0.92% -> 96.71±0.40%; 95.44±0.89% -> 96.67±0.37% | Helped |
| learn the graph jointly with graph convolution instead of using a fixed graph | GCN · MNIST, 1000 labeled | accuracy: 94.55±0.27% -> 94.28±0.28% | Hurt |
| learn the graph jointly with graph convolution instead of using a fixed graph | GCN · MNIST, 2000 and 3000 labeled | accuracy: 95.04±0.28% -> 95.09±0.17%; 95.34±0.26% -> 95.46±0.20% | Helped |
| add the graph-learning loss regularization with λ=1e-2 rather than λ=0 (ablation) | GLCN · CIFAR10 | accuracy: 0.67 -> 0.70 | Helped |
| add the graph-learning loss regularization with λ=1e-2 rather than λ=0 (ablation) | GLCN · MNIST | accuracy: 0.92 -> 0.94 | Helped |
| vary graph-learning loss weight λ across 0, 1e-4, 1e-3, 1e-2, 1e-1 and 1.0 (ablation) | GLCN · CIFAR10 and MNIST | accuracy: CIFAR10: 0.67, 0.69, 0.69, 0.70, 0.69, 0.69; MNIST: 0.92, 0.93, 0.93, 0.94, 0.94, 0.93 | Mixed |
| vary convolution-layer hidden units from 50 to 90 (ablation) | GLCN · MNIST | accuracy: 0.9410, 0.9396, 0.9394, 0.9410, 0.9389 | No clear effect |
| vary the number of convolution layers (ablation) | GLCN · MNIST | accuracy | No clear effect |
Closest papers by meaning
- Deeper Insights Into Graph Convolutional Networks for Semi-Supervised Learning (AAAI 2018 · similarity 0.94)
- Semi-Supervised Classification with Graph Convolutional Networks (ICLR 2017 · similarity 0.93)
- Contrastive and Generative Graph Convolutional Networks for Graph-based Semi-Supervised Learning (AAAI 2021 · similarity 0.92)
- L2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks (CVPR 2020 · similarity 0.91)
- Collaborative Graph Convolutional Networks: Unsupervised Learning Meets Semi-Supervised Learning (AAAI 2020 · similarity 0.91)
- 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.