Active-Bayesian learning for cooperation connectivity in dynamic cyber-physical-human systems

This work presents a novel learning framework for instrumental utility of behavioral connectivity within dynamic social networks. The combination of active and Bayesian learning techniques enables the proposed algorithm to learn how to predict a likelihood of group-stages of cooperation, according to observable behavioral connectivity within a dynamic social network. A labeler of social network data defines the group-stages of cooperation, and classifies observed behavioral connectivity according to them, in order to populate a training data pool for the proposed learning algorithm. Moreover, a multiple ordinary least squares (OLS) regression, is used to estimate the behavioral intention of social actors, in order to automate interventions for the improvement of actors' behavioral utility within the dynamic social network. We illustrate our framework through simulation examples that showcase the efficiency of the proposed algorithm to accurately predict predefined group-stages of cooperation in human-driven environments.