Reproduce “Tensorized Hypergraph Neural Networks” 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
Maolin Wang, Yaoming Zhen, Yu Pan, Yao Zhao, Chenyi Zhuang, Zenglin Xu, Ruocheng Guo, Xiangyu Zhao
A novel Tensorized Hypergraph Neural Network (THNN) directly models high-order interactions in hypergraphs using adjacency tensors and a high-order outer product message passing mechanism. The model, employing partially symmetric CP decomposition to ensure computational feasibility, demonstrates superior performance in 3D object classification, improving upon existing hypergraph neural networks.
Key insights
- Directly modeling high-order interactions with adjacency tensors and outer product feature aggregation faithfully captures complex hypergraph structures, avoiding information loss from first-order approximations.
- Tensor decomposition, specifically partially symmetric CP decomposition, effectively mitigates the "curse of dimensions" in high-order tensor operations, enabling practical deployment of such models.
- For non-uniform hypergraphs, processing hyperedges of different orders independently (THNN-Multi) generally outperforms artificially unifying them with a global node, providing more robust and order-specific feature learning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.