Research map / Graph learning / Graph self-supervised learning
Iterative Deep Graph Learning for Graph Neural Networks: Better and Robust Node Embeddings
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 |
|---|---|---|---|
| Iteratively refine graph structure and node embeddings rather than iterate only once (ablation) | IDGL and IDGL-Anch · Cora, Citeseer, Wine, Cancer, Digits, 20News | accuracy: IDGL: 84.5 (0.3) vs 83.5 (0.6); 74.1 (0.2) vs 71.0 (0.8); 97.8 (0.6) vs 97.2 (0.8); 95.1 (1.0) vs 94.7 (0.9); 93.1 (0.5) | Helped |
| Add graph regularization (ablation) | IDGL and IDGL-Anch · Cora, Citeseer, Wine, Cancer, Digits, 20News | accuracy: IDGL: 84.5 (0.3) vs 84.3 (0.4); 74.1 (0.2) vs 71.5 (0.9); 97.8 (0.6) vs 97.3 (0.8); 95.1 (1.0) vs 94.9 (1.0); 93.1 (0.5) | Helped |
| Use IDGL instead of the strongest listed baseline | IDGL · Wine | accuracy: 97.8 (0.6) vs 97.3 (0.4) | Helped |
| Use IDGL instead of the strongest listed baseline | IDGL · Cancer | accuracy: 95.1 (1.0) vs 93.4 (2.4) | Helped |
| Use IDGL instead of the strongest listed baseline | IDGL · 20News and MRD | accuracy / R²: 20News: 83.6 (0.4) vs 81.3 (0.6); MRD: 63.7 (1.8) vs 60.1 (1.5) | Helped |
| Use IDGL-Anch instead of the strongest listed baseline | IDGL-Anch · Pubmed and ogbn-arxiv | accuracy: Pubmed: 83.0 (0.2) vs 79.8 (0.4); ogbn-arxiv: 72.0 (0.3) vs 71.7 (0.3) | Helped |
| Use IDGL-Anch instead of IDGL | IDGL-Anch · Cora and Citeseer | accuracy: Cora: 84.4 (0.2) vs 84.5 (0.3); Citeseer: 72.0 (1.0) vs 74.1 (0.2) | Mixed |
| Use more anchors in IDGL-Anch | IDGL-Anch · Cora and Pubmed | accuracy: Cora: 38.3 (25.9) at 100; 58.7 (30.5) at 400; 84.4 (0.2) at 1,000. Pubmed: 82.4 (0.5) at 100; 83.0 (0.4) at 1,300 | Mixed |
| Reduce the initial-graph weight λ from 0.9 to 0.8 | IDGL · Cora | accuracy: 83.6 (0.4) -> 84.5 (0.3) | Helped |
| Reduce the initial-graph weight λ from 0.8 to 0.5 | IDGL-Anch · Cora | accuracy: 84.4 (0.2) -> 54.6 (32.3) | Hurt |
| Dynamically stop iterations using the convergence criterion instead of a fixed global iteration count | IDGL · Cora and Citeseer | accuracy | Helped |
| Lower reliance on the initial graph in edge-addition attacks | IDGL · Cora with random edge additions | accuracy | Helped |
Closest papers by meaning
- Rethinking Graph Regularization for Graph Neural Networks (AAAI 2021 · similarity 0.90)
- Beyond Graph Convolutional Network: An Interpretable Regularizer-Centered Optimization Framework (AAAI 2023 · similarity 0.89)
- Distance Encoding: Design Provably More Powerful Neural Networks for Graph Representation Learning (NeurIPS 2020 · similarity 0.89)
- GraphGAN: Graph Representation Learning With Generative Adversarial Nets (AAAI 2018 · similarity 0.89)
- AdaGCN: Adaboosting Graph Convolutional Networks into Deep Models (ICLR 2021 · similarity 0.89)
- Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · similarity 0.88)
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.