We present an autonomous adaptive quality of service (QoS) driven network system called a "cognitive packet network" (CPN), which adaptively selects paths so as to offer best effort QoS to the end users based on user defined QoS. CPN uses neural network based reinforcement learning to make routing decisions separately at each node. Measurements on an experimental test-bed are provided to show how the system responds to the choice of QoS goals. We also discuss and evaluate an extension of CPN that uses a genetic algorithm to generate and maintain paths from previously discovered information by matching their "fitness" with respect to the desired QoS.
Available at: http://works.bepress.com/peixiang-liu/3/