Approximating behavioral equivalence for scaling solutions of I-DIDs. Issue 2 (November 2016)

• Record Type: Journal Article
• Authors: Zeng, Yifeng; Doshi, Prashant; Chen, Yingke; Pan, Yinghui; Mao, Hua; Chandrasekaran, Muthukumaran
• Is Part Of: Knowledge and information systems. Volume 49:Issue 2(2016:Nov.)
• Page Start: 511
• Page End: 552
• Publication Date: 2016-11
• Subjects: Multiagent systems -- Decision making -- Influence diagrams -- Behavioral equivalence; Expert systems (Computer science) -- Periodicals; Information storage and retrieval systems -- Periodicals; 006.33
• DOI: 10.1007/s10115-015-0912-x
• Languages: English
• ISSNs: 0219-1377

Abstract

Abstract Interactive dynamic influence diagram (I-DID) is a recognized graphical framework for sequential multiagent decision making under uncertainty. I-DIDs concisely represent the problem of how an individual agent should act in an uncertain environment shared with others of unknown types. I-DIDs face the challenge of solving a large number of models that are ascribed to other agents. A known method for solving I-DIDs is to group models of other agents that are behaviorally equivalent. Identifying model equivalence requires solving models and comparing their solutions generally represented as policy trees. Because the trees grow exponentially with the number of decision time steps, comparing entire policy trees becomes intractable, thereby limiting the scalability of previous I-DID techniques. In this article, our specific approaches focus on utilizing partial policy trees for comparison and determining the distance between updated beliefs at the leaves of the trees. We propose a principled way to determine how much of the policy trees to consider, which trades off solution quality for efficiency. We further improve on this technique by allowing the partial policy trees to have paths of differing lengths. We evaluate these approaches in multiple problem domains and demonstrate significantly improved scalability over previous approaches.