Research map / Graph learning / GNN architectures
Label Deconvolution for Node Representation Learning on Large-Scale Attributed Graphs Against Learning Bias
Zhihao Shi, Jie Wang, Fanghua Lu, Hanzhu Chen, Defu Lian, Zheng Wang, Jieping Ye, Feng He Wu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 inverse labels (label deconvolution) instead of pseudo labels on true labels only (ablation) | fine-tuned node encoder · ogbn-product | accuracy: 83.83 -> 86.45 | Improved |
| add inverse labels instead of pseudo labels (ablation) | fine-tuned node encoder · ogbn-arxiv | accuracy: 75.35 -> 77.26 | Improved |
| add inverse labels instead of pseudo labels (ablation) | fine-tuned node encoder · ogbn-protein | ROC-AUC: 89.15 -> 89.42 | Improved |
| add pseudo labels on evaluation nodes to true-label fine-tuning (ablation) | X_FNE · ogbn-product | accuracy: 82.88 -> 83.83 | Improved |
| add pseudo labels on evaluation nodes to true-label fine-tuning (ablation) | X_FNE · ogbn-arxiv | accuracy: 75.52 -> 75.35 | No clear effect |
| replace GLEM with LD (single fine-tuning) | separate training · ogbn-product | accuracy: 85.09 -> 86.45 | Improved |
| replace GLEM with LD | separate training · ogbn-product | training time: 75.67h -> 24.78h | Improved |
| use LD instead of fine-tuning with true labels | X_FNE · ogbn-product | accuracy: 82.88 -> 86.45 | Improved |
| use LD instead of fine-tuning with true labels | X_FNE · ogbn-product | training time: 16.88h -> 24.78h | Hurt |
| use LD in inductive setting vs pre-trained embeddings | X_PNE · ogbn-arxiv inductive | accuracy: 75.23 -> 76.83 | Improved |
| use LD in inductive setting vs GLEM | X_GLEM · ogbn-protein inductive | ROC-AUC: 88.72 -> 89.12 | Improved |
| shrink sampled subgraph size for joint training | GCN+GAS · ogbn-arxiv | accuracy: 74.28% -> 67.46% | Hurt |
Closest papers by meaning
- Active Learning on Attributed Graphs via Graph Cognizant Logistic Regression and Preemptive Query Generation (ICML 2020 · similarity 0.90)
- Label-free Node Classification on Graphs with Large Language Models (LLMs) (ICLR 2024 · similarity 0.89)
- Inductive Representation Learning on Large Graphs (NeurIPS 2017 · similarity 0.89)
- Subgraph Invariant Learning Towards Large-Scale Graph Node Classification (AAAI 2025 · similarity 0.88)
- Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning (ICLR 2025 · similarity 0.88)
- NLGT: Neighborhood-based and Label-enhanced Graph Transformer Framework for Node Classification (AAAI 2025 · 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.
