Research map / Label-efficient and robust learning / Domain adaptation and generalization
When Sparse Graph Representation Learning Falls into Domain Shift: Data Augmentation for Cross-Domain Graph Meta-Learning (Student Abstract)
Read the paperCheck your idea against 108,967 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 CDA-GML, a cross-domain graph meta-learning method with adversarial-gradient data augmentation and pseudo-unseen domain simulation, evaluated for few-shot node classification on AMiner (System domain).
Problem
Graph meta-learning generalizes poorly to out-of-distribution domains under label sparsity and domain shift, especially when no target-domain data is available for adaptation.
Approach
Combines cross-domain graph meta-learning, which simulates domain shift by meta-testing on pseudo-unseen tasks, with a data augmentation module that perturbs node embeddings using adversarial gradients from the CGML classifier, so no target data is needed.
Key results
- CDA-GML reaches 85.76% on 2-way 1-shot System-domain node classification, versus 83.81% for CrossHG-Meta (best baseline).
- On 3-way 3-shot, CDA-GML achieves 86.68% versus 84.57% for CrossHG-Meta.
- Ablations show removing either the CGML module (\CD) or the data augmentation module (\DA) lowers accuracy in all settings.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use CDA-GML instead of CrossHG-Meta | Cross-domain few-shot node classification · AMiner System, 2-way 1-shot | accuracy (%): 83.81 -> 85.76 | Helped |
| use CDA-GML instead of CrossHG-Meta | Cross-domain few-shot node classification · AMiner System, 2-way 3-shot | accuracy (%): 91.44 -> 93.83 | Helped |
| use CDA-GML instead of CrossHG-Meta | Cross-domain few-shot node classification · AMiner System, 3-way 1-shot | accuracy (%): 75.59 -> 78.21 | Helped |
| use CDA-GML instead of CrossHG-Meta | Cross-domain few-shot node classification · AMiner System, 3-way 3-shot | accuracy (%): 84.57 -> 86.68 | Helped |
| add the CGML module (ablation) | CDA-GML · AMiner System, 2-way 1-shot | accuracy (%): 84.36 -> 85.76 | Helped |
| add the Data Augmentation module (ablation) | CDA-GML · AMiner System, 2-way 1-shot | accuracy (%): 84.19 -> 85.76 | Helped |
| add the CGML module (ablation) | CDA-GML · AMiner System, 2-way 3-shot | accuracy (%): 92.25 -> 93.83 | Helped |
| add the Data Augmentation module (ablation) | CDA-GML · AMiner System, 2-way 3-shot | accuracy (%): 92.16 -> 93.83 | Helped |
| add the CGML module (ablation) | CDA-GML · AMiner System, 3-way 1-shot | accuracy (%): 77.14 -> 78.21 | Helped |
| add the Data Augmentation module (ablation) | CDA-GML · AMiner System, 3-way 1-shot | accuracy (%): 76.70 -> 78.21 | Helped |
| add the CGML module (ablation) | CDA-GML · AMiner System, 3-way 3-shot | accuracy (%): 85.49 -> 86.68 | Helped |
| add the Data Augmentation module (ablation) | CDA-GML · AMiner System, 3-way 3-shot | accuracy (%): 85.22 -> 86.68 | Helped |
Limitations
- Evaluated only on a single target domain (System) of the AMiner dataset with small-scale few-shot settings.
- Method description is brief; the augmentation formula and the adversarial gradient computation lack full detail.
- Reported gains are modest and no variance or significance statistics are given.
Experimental setup
Datasets: AMiner
Benchmarks: AMiner
Metrics: node classification accuracy (%)
Baselines: GCN R-GCN ProtoNet MAML Meta-SGC AMM-GNN CrossHG-Meta
Builds on: MAML CrossHG-Meta
Closest papers by meaning
- Context-aware Graph Meta-learning (AAAI 2026 · similarity 0.90)
- Multi-Domain Generalized Graph Meta Learning (AAAI 2023 · similarity 0.89)
- Graph Meta Learning via Local Subgraphs (NeurIPS 2020 · similarity 0.89)
- Cross-View Graph Consistency Learning for Invariant Graph Representations (AAAI 2025 · similarity 0.85)
- On the Benefits of Attribute-Driven Graph Domain Adaptation (ICLR 2025 · similarity 0.85)
- Cross-Domain Graph Data Scaling: A Showcase with Diffusion Models (NeurIPS 2025 · similarity 0.85)
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.