SCGNN: Efficient Graph Learning via Granular-ball Computing
A new paper on arXiv (2605.02617v2) introduces SCGNN, a plug-and-play framework for graph neural networks that uses granular-ball computing to capture semantic consistency among nodes. Unlike traditional methods that rely on k-nearest neighbors or full search algorithms with high computational cost and noise, SCGNN adaptively partitions nodes into granular balls to model group-level structure, reducing complexity and improving robustness. The approach is designed to be scalable and noise-resistant.
Key facts
- arXiv paper 2605.02617v2
- SCGNN stands for Semantic Consistency enhanced Graph Neural Network
- Uses granular-ball computing (GBC)
- Replaces k-nearest neighbors and full search algorithms
- Reduces computational complexity
- Improves robustness to noise
- Plug-and-play framework
- Models group-level semantic structure
Entities
Institutions
- arXiv