Dynamic Hypergraph Model Learns from Time Series Without Prior Knowledge
A new model constructs dynamic hypergraph representations from multivariate time series data without requiring prior knowledge of the underlying structure. The approach applies community detection to time series, uses an attention mechanism to transform communities into hypergraphs via a clique-based technique. The resulting hypergraphs are then processed by a Dynamic Hypergraph Attention Convolution network. This work addresses the challenge of deriving hypergraph representations when the hypergraph structure is limited or absent, enabling the capture of higher-dimensional relationships in complex systems. The study is published on arXiv as preprint 2605.22540.
Key facts
- Model constructs dynamic hypergraph from multivariate time series without prior knowledge.
- Uses community detection and attention mechanism.
- Transforms communities into hypergraphs via clique-based technique.
- Hypergraphs processed by Dynamic Hypergraph Attention Convolution.
- Published on arXiv with ID 2605.22540.
- Addresses challenge of deriving hypergraph representations from time series.
- Captures higher-dimensional relationships in complex systems.
- No prior knowledge of hypergraph structure required.
Entities
Institutions
- arXiv