- By:
- Ahmed taimoor, Syed; Shivakumar, Shruti; Liu, Xiaorui; Kannan, Ramakrishnan ; Li, Jiajia
- Page Number:
- 6676-6680
- Book Title:
- 2025 IEEE International Conference on Big Data (BigData)
- Publication Date:
- September 15, 2026
- Publisher Location:
- IEEE, New Jersey, United States of America
- Conference Name:
- The 13th IEEE International Conference on Big Data (IEEE BigData 2025)
- Conference Location:
- Macau, China
- Conference Sponsor:
- Institute of Electrical and Electronics Engineers
- View DOI Listing:
- https://doi.org/10.1109/BigData66926.2025.11401757
Abstract
Many forms of real-world data across diverse domains encode relationships among multiple objects, with networks and point cloud data being common examples. Hypergraphs naturally capture such higher-order relations, and Hypergraph Neural Networks (HGNNs) have been shown to outperform their graph-based counterparts (GNNs) on many tasks. Some recent HGNN architectures leverage tensor-based representations that improve task performance over matrix-based approaches but incur significantly higher computational and memory costs. In this work, we present a scalable and efficient tensor-based HGNN (ST-HGNN), using T-MPHN as an archetype. We address two key bottlenecks of tensor-based HGNNs: (1) efficient tensor representation generation, and (2) replacing the recursive message-passing algorithm with an iterative implementation. Our optimizations yield substantial improvements in scalability and runtime performance, achieving up to 61.46× speed-up for data structure generation and 54.61× for message passing.
