Cognition on the networked data of stochastic topology

In this paper we address the problem of cognition (inference and decision) on data collected from a system which is characterized by a stochastic network topology. The decision problem is formulated as a data fusion problem and a data model that only requires stochastic network topology, marginal probability densities and pairwise correlations is proposed. Based on such model, we can mitigate the difficulties of inference resulted from random topological networks and simplify the decision making in many data cases. The numerical simulation justifies our idea and the performance is comparable to the case of decision making with complete information and decision making under conventional independent data cases when the size of network is not too large. Furthermore, an experiment of classification of real-world data is also conducted to illustrate the potential applications of cognition of networked data on social networks.