Bayesian constrained decision fusion
Panagiotis A. Traganitis, Georgios B. Giannakis · 2021
Decision fusion aims to intelligently combine decisions provided by a network of sensors. However, uncalibrated sensors or sensors of unknown reliability challenge this task because they significantly skew the fused decision. This work deals with decision fusion when no information on the sensor reliability is provided. To ensure high-performance fusion, side information is leveraged in the form of pairwise constraints, that capture relationships between pairs of data. A Bayesian approach is developed based on variational inference that can jointly assess sensor reliability, and perform label aggregation. Performance of the proposed algorithm is validated on real datasets.