Reproduce “Jointly learning relevant subgraph patterns and nonlinear models of their indicators” 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
Ryo Shirakawa, Yusei Yokoyama, Fumiya Okazaki, Ichigaku Takigawa
An algorithm is developed that integrates Gradient Tree Boosting with efficient subgraph pattern mining, enabling the learning of nonlinear predictive models from graph-structured data. It achieves 100% accuracy on a synthetic Graph-XOR problem requiring nonlinearity and often outperforms linear models on real-world QSAR datasets.
Key insights
- The hypothesis space of nonlinear models of connected subgraph indicators is strictly larger than that of linear models, indicating the necessity of nonlinearity for certain graph problems.
- Integrating graph pattern mining directly into a powerful nonlinear learning framework like GTB allows for simultaneous, efficient feature discovery and model building.
- The proposed pruning strategy significantly improves computational efficiency by systematically avoiding the exploration of subgraph extensions that cannot yield a better data split.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.