Research map / Graph learning / Graph self-supervised learning
MDGMIX: Boundary-Aware Subgraph Mixing for Multi-Domain Graph Pre-Training
Ziyu Zheng, Yaming Yang, Ziyu Guan, Wei Zhao, Xinyan Huang
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 boundary-aware subgraph mixing and hierarchical discrimination to multi-domain graph pre-training | MDGMIX · 1-shot node classification, 8 target graphs | accuracy: 43.38 -> 46.83 (Cora vs MDGPT) | Improved |
| use boundary node selection instead of random subgraph mixup (ablation) | MDGMIX · Cora node classification | accuracy: 45.01 -> 46.83 | Improved |
| add coarse-grained domain classification loss (ablation) | MDGMIX · Cora node classification | accuracy: 44.22 -> 46.83 | Improved |
| add fine-grained domain decomposition loss (ablation) | MDGMIX · Cora node classification | accuracy: 43.15 -> 46.83 | Improved |
| add boundary mixup and domain decomposition together over domain classification alone (ablation) | MDGMIX · Photo node classification | accuracy: 62.79 -> 68.40 | Improved |
| keep domain decomposition when removing boundary mixup and domain classification (ablation) | MDGMIX Variant 4 · Cora node classification | accuracy: 45.28 vs 46.83 full | Improved |
| mix more than two domains via Dirichlet weights (ablation) | MDGMIX · 7 target graphs | accuracy: Cora 46.20, 45.41, 45.06, 45.47 for 3-6 domains | No clear effect |
| sample lambda from Beta distribution instead of fixed 0.5 (ablation) | MDGMIX · Cora, Computers, Chameleon | accuracy: Computers 53.81 (Beta(0.2,0.2)) -> 54.95 (fixed 0.5) | No clear effect |
| randomly drop up to 90% of nodes from pre-training graphs (ablation) | multi-domain graph pre-training baselines · Pubmed, Computers | Micro accuracy | No clear effect |
| pre-train on boundary-centered mixed subgraphs and adapt with prompt weighting instead of full source graphs and domain tokens | MDGMIX · pre-training and fine-tuning efficiency | time, memory, trainable parameters: fine-tuning parameters 6 vs 59 to 10240 for baselines | Improved |
| use MDGMIX on boundary subgraphs versus center or random subgraphs for domain classification (ablation) | domain classifier · boundary analysis | domain classification accuracy | Improved |
Closest papers by meaning
- How Much Can Transfer? BRIDGE: Bounded Multi-Domain Graph Foundation Model with Generalization Guarantees (ICML 2025 · similarity 0.89)
- Unified Graph Neural Networks Pre-training for Multi-domain Graphs (AAAI 2025 · similarity 0.88)
- GraphKeeper: Graph Domain-Incremental Learning via Knowledge Disentanglement and Preservation (NeurIPS 2025 · similarity 0.87)
- Multi-Domain Graph Foundation Models: Robust Knowledge Transfer via Topology Alignment (ICML 2025 · similarity 0.87)
- Towards Unsupervised Open-Set Graph Domain Adaptation via Dual Reprogramming (NeurIPS 2025 · similarity 0.87)
- Open-Set Graph Domain Adaptation via Separate Domain Alignment (AAAI 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.
