What are Graphical Models for Multiagent Decision Making?

Interactive Dynamic Influence Diagram
An influence diagram (ID) (also called a relevance diagram, decision diagram or a decision network) is a compact graphical and mathematical representation of a decision situation. It is a generalization of a Bayesian network, in which not only probabilistic inference problems but also decision making problems (following maximum expected utility criterion) can be modeled and solved. Their dynamic counterparts, Dynamic Influence Diagrams (DIDs) are graphical representations for partially observable Markov decision processes (POMDPs). Researchers developed new graphical representations for the problem of sequential decision making in partially observable 'multiagent' environments, as formalized by interactive partially observable Markov decision processes (IPOMDPs). The graphical models called interactive influence diagrams (IIDs) and their dynamic counterparts, interactive dynamic influence diagrams (IDIDs), seek to explicitly model the structure that is often present in realworld problems by decomposing the situation into chance and decision variables, and the dependencies between the variables. IDIDs generalize DIDs to multiagent settings in the same way that IPOMDPs generalize POMDPs. IDIDs may be used to compute the policy of an agent given its belief as the agent acts and observes in a setting that is populated by other interacting agents. These graphical models are significantly more transparent and semantically clear than other previous representations.


Project Description

IDIDs closely model realistic partially observable multiagent scenarios that are commonly observed in the real world. In such scenarios, the subject agent not only models the uncertainties that may exist in the environment it is in, but also models the uncertainties that may exist in how the other agents behave. While attempting to capture this, algorithms for solving IDIDs face the challenge of an exponentially growing space of candidate models ascribed to other agents, over time. We seek to develop efficient exact and approximate methods to efficiently solve them thereby allowing scalability to larger horizons as well as larger problems with more number of states.


Project Generated Resources

Netus v1.0
 Opensource Interactive dynamic influence diagram (IDID) source and executable files available for download and use under Affero GNU public license v3.0 ReadMe , Source Code
MAIDSolver v1.0
 Opensource multiagent influence diagram (MAID) solver available for download and use under Affero GNU public license v3.0 ReadMe , Source Code
Publications:
 Yifeng Zeng, Prashant Doshi, Yinghui Pan, Hua Mao and Muthukumaran Chandrasekaran, "Utilizing Partial Policies for Identifying Equivalence of Behavioral Models", in AAMAS, 2011
 Prashant Doshi, Muthukumaran Chandrasekaran and Yifeng Zeng, "EpsilonSubjective Equivalence of Models for Interactive Dynamic Influence Diagrams", in WIC/IEEE/ACM IAT 2010
 Muthukumaran Chandrasekaran, Prashant Doshi and Yifeng Zeng, "Approximate Solutions of Interactive Dynamic Influence Diagrams Using epsilonBehavioral Equilvance", in ISAIM 2010
 Yifeng Zeng and Prashant Doshi, "Speeding Up Exact Solutions of Interactive Dynamic Influence Diagrams Using Action Equivalence", in IJCAI, 2009 Talk
 Prashant Doshi and Yifeng Zeng, "Improved Approximation of Interactive Dynamic Influence Diagrams Using Discriminative Model Updates", in AAMAS, 2009 Talk Poster
 Yifeng Zeng and Prashant Doshi, "Model Identification in Interactive Influence Diagrams Using Mutual Information", in Journal of Web Intelligence and Agent Systems (WIAS), IOS Press, 8(3):313327, 2010.
 Yifeng Zeng and Prashant Doshi, "An Informationtheoretic Approach to Model Identification in Interactive Influence Diagrams", in IEEE/WIC/ACM IAT, 2008 Talk
 Yifeng Zeng and Prashant Doshi, "Toward Robust Model Identification in Interactive Influence Diagrams Using Mutual Information", in MSDM Workshop, AAMAS, 2008 Talk
 Prashant Doshi and Yifeng Zeng, "Graphical Models for Interactive POMDPs: Representations and Solutions", in Journal of Autonomous Agents and Multiagent Systems (JAAMAS), 18(3):376416, 2009
 Yifeng Zeng, Prashant Doshi, and Qiongyu Chen, "Approximate Solutions to Dynamic Interactive Influence Diagrams Using Model Clustering", in AAAI, 2007 Talk
 Prashant Doshi, Yifeng Zeng, and Qiongyu Chen, "Graphical Models for Online Solutions to Interactive POMDPs", in AAMAS, 2007 Talk
 Prashant Doshi, Yifeng Zeng, and Qiongyu Chen, "Graphical Models for Online Solutions to Interactive POMDPs", in AAAI Spring Symposium on GTDT Agents, 2007




Collaborating Institutions



Researchers

 Prashant Doshi (PI), THINC Lab, UGA
 Yifeng Zeng (Associate PI), Teesside University
 Muthukumaran Chandrasekaran, PhD Student, THINC Lab, UGA
 Fadel Adoe, PhD Student, THINC Lab, UGA
 Yingke Chen, Post Doctoral Student, THINC Lab, UGA


