Unified Multiagent Markov Models’ Complexity Analysis

In many domains and particularly in multiagent problems, the characterization of the information available to each agent about her environment impacts directly on the problem complexity. In this paper, we explore how influence diagrams can help in understanding precisely the impact of the structure of markovian multiagent models on their theoretical complexity. More precisely, we use two different classes of algorithms (a variable elimination algorithm and a tree search based method) to bound the complexity of such models in a unified manner. We finally discuss about theoretical results of this work and further research avenues.