Research map / Graph learning / Graph self-supervised learning
Graph is a Natural Regularization: Revisiting Vector Quantization for Graph Representation Learning
Zian Zhai, Fan Li, Xingyu Tan, Xiaoyang Wang, Wenjie Zhang
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 RGVQ (Gumbel-Softmax plus structure-aware regularization) to Graph VQ | GFT · Cora codebook perplexity (K=512) | perplexity: 94.47 -> 211.69 | Improved |
| add RGVQ | GFT · PubMed codebook perplexity (K=512) | perplexity: 4.14 -> 319.09 | Improved |
| add RGVQ | GFT · Average over 7 pretrain-finetune datasets | accuracy / ROC-AUC: 79.50 -> 81.42 | Improved |
| add RGVQ | GQT · Cora node classification | accuracy: 86.44 -> 88.34 | Improved |
| add RGVQ | GQT · Roman node classification | accuracy: 89.85 -> 90.98 | Improved |
| add SimVQ collapse mitigation | GFT · Average over 7 datasets | accuracy / ROC-AUC: 79.50 -> 75.97 | Hurt |
| add EMA codebook update | GFT · Average over 7 datasets | accuracy / ROC-AUC: 79.50 -> 79.73 | No clear effect |
| add reparameterization with structural regularization only (remove feature signal) (ablation) | RGVQ · Cora | accuracy: 78.35 -> 79.87 | Improved |
| add both structural and feature regularization versus structure only (ablation) | RGVQ · PubMed | accuracy: 76.32 -> 77.46 | Improved |
| remove Gumbel-Softmax reparameterization (ablation) | RGVQ · Cora/PubMed/WikiCS | perplexity: 319.09 -> 4.14 (PubMed) | Hurt |
| lower Gumbel-Softmax temperature (ablation) | RGVQ · codebook perplexity | perplexity | Improved |
| increase number of contrastive samples (ablation) | RGVQ · codebook perplexity | perplexity | No clear effect |
Closest papers by meaning
- Addressing Representation Collapse in Vector Quantized Models with One Linear Layer (ICCV 2025 · similarity 0.89)
- CauVQ: Causal Vector Quantization for Graph OOD Generalization (AAAI 2026 · similarity 0.88)
- Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · similarity 0.87)
- VQGraph: Rethinking Graph Representation Space for Bridging GNNs and MLPs (ICLR 2024 · similarity 0.87)
- Outlier-Aware Post-Training Quantization for Discrete Graph Diffusion Models (ICML 2025 · similarity 0.86)
- VCR-Graphormer: A Mini-batch Graph Transformer via Virtual Connections (ICLR 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.
