Toward data-driven solutions to interactive dynamic influence diagrams

Abstract With the availability of significant amount data, data-driven decision making becomes an alternative way for solving complex multiagent problems. Instead using domain knowledge to explicitly build models, approach learns decisions (probably optimal ones) from available data. This removes bottleneck in traditional knowledge-driven making, which requires a strong support experts. In this paper, we study context interactive dynamic influence diagrams (I-DIDs)—a general framework sequential under uncertainty. We propose solve I-DIDs model and focus on learning behavior other agents problem domains. The challenge is complete policy tree that will be embedded models due limited two new methods develop trees I-DIDs. first method uses simple clustering process, while second one employs sophisticated statistical checks. analyze proposed algorithms theoretical experiment them over