Chinese Journal of Computers   Full Text
Title	Research on Multi-Agent Dynamic Influence Diagrams and Its Approximate Inference Algorithm
Authors	YAO Hong-Liang WANG Hao ZHANG You-Sheng WANG Rong-Gui
Address	(Department of Computer Science and Technology, Hefei University of Technology, Hefei 230009)3
Year	2008
Issue	No.2(236—244)
Abstract &   Background	Abstract As MAIDs can not model dynamic environment and Multi-Agent MDPs are difficult to represent structural relations among Agents, Multi-Agent dynamic influences diagrams(MADIDs) are presented for modeling structural relations of Multi-Agent system in dynamic environment. For efficiently inferring in MADIDs, an extensional BK(EBK) algorithm is proposed, and the extensions are in three aspects: The marginalizing operation of utility nodes is added in BK algorithm； the separators are added for decreasing the error of inference； the junction tree that obtained by hierarchical decomposition of MADIDs is used for improving the efficiency of inference. Given model instances, the experiment results show the validity of MADIDs and EBK algorithm. keywords multi-Agent influence diagrams; multi-Agent dynamic influence diagrams; junction tree; BK algorithm background Due to high dimension and uncertainty of the complex system, the complexity system is often hard to be represented and processed. The knowledge representation and computation methods of complex system are open hard problems in complex system research. The complex intelligent system is an important branch of the complexity system. At present, many the researchers use probability models to model and process the complexity intelligent system, such as Multi-Agent influence diagrams(MAIDs) and Multi-Agent Markov decision processes(MMDPs). In order to better represent and process complex dynamic intelligent system, the authors of the paper present a new probability model, which is Multi-Agent Dynamic Influence Diagrams(MADIDs). Another, an approximate inference algorithm of MADIDs is proposed.