Reproduce “ROLAND: Graph Learning Framework for Dynamic Graphs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiaxuan You, Tianyu Du, Jure Leskovec
The ROLAND framework from Stanford University adapts static Graph Neural Networks for dynamic graphs, introducing a live-update evaluation and incremental training strategy. It achieves an average relative MRR improvement of 15.5% over adapted baselines in live-update settings and scales to datasets with 56 million edges, overcoming memory limitations.
Key insights
- Existing static GNN architectural components (e.g., batch normalization, skip-connections, max aggregation) are crucial for improving dynamic GNN performance.
- A 'live-update' evaluation paradigm is essential for accurately benchmarking dynamic GNNs, revealing the limitations of traditional static temporal splits.
- Incremental training significantly enhances scalability, allowing dynamic GNNs to handle much larger datasets than previously possible with back-propagation-through-time, while meta-learning improves adaptation speed and stability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.