A New Stochastic Algorithm For The Capacity & Flow Assignment Problem And An Associated Rate Of Convergence

In this paper we examine the problem of optimizing the average time latency of a network using agents that are able to learn. The network design is constrained by a traffic matrix which dedicates specific flows between specific pairs of nodes. Although this is an analysis of an application, we only present two methodologies here, i.e. an algorithm for optimization and a corresponding conservative rate of convergence based on no learning. The application part will be presented in the near future once data is available. We expect the tools developed in this paper can be used to optimize a wide range of objective functions. They will not be limited to optimizing time latency.