Decentralized probabilistic reasoning, constraint reasoning, and decision theoretic reasoning are some of the essential tasks of a multiagent system (MAS). Many frameworks exist for these tasks, and a number of them organize agents into a junction tree (JT). Although these frameworks all reap benefits of communication efficiency and inferential soundness from the JT organization, their potential capacity on agent privacy has not been realized fully. The contribution of this work is a general approach to construct the JT organization through a maximum spanning tree (MST), and a new distributed MST algorithm, that preserve agent privacy on private variables, shared variables and agent identities.
展开▼