Research map / Graph learning / Graph self-supervised learning
Robust Graph Diffusion Model
Yancheng Wang, Ping Li, Dongfang Sun, Yingzhen Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a diffusion-based approach to synthesize graph structure (synthetic nodes and their edges) within an existing graph, as stated in the abstract.
Problem
Existing diffusion models for graphs generate whole or small graphs, but no diffusion approach synthesizes graph structures inside an existing graph, including synthetic nodes and their connections.
Approach
Applies diffusion generation to extend an existing graph with synthetic nodes and edges, rather than generating entire graphs from scratch. Only the abstract's framing is available, so finer details are not stated.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| augment training graph with RGDM synthetic structures | GCN · Cora, 25% Metattack | accuracy: 47.5 -> 61.4 | Improved |
| augment training graph with RGDM synthetic structures | SG-GSR · Cora, 25% Metattack | accuracy: 77.8 -> 79.1 | Improved |
| augment clean graph with RGDM | GCN and SG-GSR · clean graphs (4 datasets) | accuracy: +0.4 to +2.6 for SG-GSR | Improved |
| add sparsification loss to RGAE (ablation) | RGDM (GCN) · Cora, 25% Metattack | accuracy: 56.72 -> 61.42 | Improved |
| add low-rank latent regularization (TNN) (ablation) | RGDM (GCN) · Cora, 25% Metattack | accuracy: 60.04 -> 61.42 | Improved |
| use TNN instead of weight orthogonality or low-rank adjacency regularizers (ablation) | RGDM (GCN) · Citeseer, 25% Metattack | accuracy: 61.12 -> 62.84 | Improved |
| add edge selection to DoG augmentation (ablation) | DoG with SG-GSR · 25% Metattack, Cora | accuracy: 75.6 -> 76.6 | Improved |
| add latent TNN (ablation) | DoG + edge selection · 25% Metattack, Cora | accuracy: 76.6 -> 77.4 | Improved |
| replace BLND decoder with SHED (ablation) | RGDM · 25% Metattack, Cora | accuracy: 77.4 -> 79.1 | Improved |
| use plain DoG augmentation with SG-GSR (ablation) | SG-GSR · 25% Metattack, Cora | accuracy: 77.8 -> 75.6 | Hurt |
| use RGDM over DoG and LR-FGDM augmentation | GCN · 25% Metattack | accuracy: +2.02 over DoG on Cora | Improved |
| use RGDM on heterophilic graphs | LHS · Chameleon, Actor under Metattack | accuracy: 70.03 -> 72.41 (Chameleon 25%) | Improved |
Closest papers by meaning
- When Edge Independence Fails: Joint Graph Diffusion with Latent Sociability Priors (NeurIPS 2026 · similarity 0.89)
- Unifying Generation and Prediction on Graphs with Latent Graph Diffusion (NeurIPS 2024 · similarity 0.88)
- A Generalized Neural Diffusion Framework on Graphs (AAAI 2024 · similarity 0.88)
- SBGD: Improving Graph Diffusion Generative Model via Stochastic Block Diffusion (ICML 2025 · similarity 0.87)
- Editing Partially Observable Networks via Graph Diffusion Models (ICML 2024 · similarity 0.87)
- Hyperbolic Graph Diffusion Model (AAAI 2024 · similarity 0.87)
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.
