Beihang University proposes clustering framework for federated trajectory data valuation
Researchers designed an algorithm that accelerates Shapley value calculation across distributed trajectory systems using cluster centers.
A research team led by Professor Yongxin Tong from Beihang University has introduced a clustering-based valuation framework designed for federated trajectory data systems, according to EurekAlert eurekalert.org
EurekAlerteurekalert.orgOpen source ↗. The framework aims to resolve data silo constraints by fairly quantifying individual participant contributions to incentivize secure data sharing among distinct organizations.
Traditional data valuation approaches face computational bottlenecks when calculating the Shapley value across large-scale distributed mobility databases. In real-world federations, trajectory datasets are kept by separate entities, and standard cooperative game-theory calculations scale poorly as participant volume and spatial-temporal points expand.
Spatial-temporal decomposition
To manage computational complexity, the system utilizes the decomposable property of the utility function across individual "road segment-time" pairs. This operation converts the global valuation workload into independent sub-units, breaking down continuous movement traces into localized components that can be analyzed independently across the federation.
Cluster-based Shapley propagation
The framework groups road segment-time pairs with similar trajectory coverage patterns using a clustering algorithm to locate structural redundancies eurekalert.org
EurekAlerteurekalert.orgOpen source ↗. The system calculates the exact Shapley value exclusively for the center element of each cluster, then propagates the output to every corresponding member within that cluster group. In tests conducted on real-world datasets, the authors reported that the clustering architecture improved computational speed and valuation accuracy compared to baseline valuation methods.
