An algorithmic approach to identify irrelevant information in sequential teams

An algorithmic framework that identifies irrelevant data (i.e., data that may be ignored without any loss of optimality) at agents of a sequential team is presented. This framework relies on capturing the properties of a sequential team that do not depend on the specifics of state spaces, the probability law, the system dynamics, or the cost functions. To capture these properties the notion of a team form is developed. A team form is then modeled as a directed acyclic graph and irrelevant data is identified using D-separation properties of specific subsets of nodes in the graph. This framework provides an algorithmic procedure for identifying and ignoring irrelevant data at agents, and thereby simplifying the form of control laws that need to be implemented.