Fabila-Carrasco, John Stewart, Lei, Mei-San Maggie, Tan, Chao and Escudero, Javier
2026.
Graph-based multivariate multiscale dispersion entropy: efficient implementation and applications to real-world network dataintegrating SAM supervision for 3D weakly supervised point cloud segmentation.
IEEE Transactions on Signal and Information Processing over Networks
12
, pp. 732-742.
10.1109/TSIPN.2026.3688396
|
Preview |
PDF
- Accepted Post-Print Version
Available under License Creative Commons Attribution. Download (669kB) | Preview |
Abstract
We introduce Graph-based Multivariate Multiscale Dispersion Entropy (mvDE_G), a novel and computationally efficient method for analysing multivariate time series on graphs. This approach significantly extends traditional nonlinear analysis techniques by unifying the temporal dimension of data with its topological relationships, captured by the graph, enabling more comprehensive analysis. Notably, mvDE_G is computationally efficient: for a dataset with 2000 samples, 8 channels, and embedding dimension 5, mvDE_G requires only ∼1.6×104 pattern evaluations compared to ∼1.3×109 for classical mvDE, a reduction of nearly five orders of magnitude. For fixed and practically relevant embedding dimensions, our optimized procedure exhibits linear growth in runtime with respect to the number of channels, whereas classical mvDE retains a combinatorial dependence on m. This is achieved through refined matrix-power computations and Kronecker products. In synthetic scenarios with correlated noise, mvDE_G successfully distinguishes various degrees of complexity, with minimal overlap in standard deviations across different correlation levels (0.15-0.95), whereas classical mvDE shows complete overlap. For real-world data, such as two-phase flow regimes, weather, and sensor network monitoring, mvDE_G identifies distinct dynamical patterns across both small and large scales, offering better discrimination than univariate or purely multivariate (no-graph) techniques. Overall, mvDE_G enables fast and scalable nonlinear analysis for time series on graphs, providing a promising tool for real-time graph data analysis in a wide range of applications.
| Item Type: | Article |
|---|---|
| Date Type: | Publication |
| Status: | Published |
| Schools: | Schools > Computer Science & Informatics |
| Additional Information: | RRS policy applied |
| Publisher: | IEEE |
| ISSN: | 2373-776X |
| Date of First Compliant Deposit: | 6 July 2026 |
| Last Modified: | 02 Aug 2026 08:30 |
| URI: | https://orca.cardiff.ac.uk/id/eprint/187213 |
Actions (repository staff only)
 |
Edit Item |
Altmetric
Altmetric