ILS:A System of Learning Distributed Heterogeneous Agents for Network Traffic Management

摘要

Our initial implementation of a domainindependent Integrated Learning System(ILs)discovers how to control the traffic in a telecommunications network through its own experience.The network traffic control problem is to“optimize”the behaviour of a network by intelligently placing controls on network elements. Network Traffic Management is of great economic importance because customers are not charged for calls that fail to complete.The problem is very data-ntensive and a number of approaches have been tried,including several in Artificial Intelligence,The hovel use of learning in this task has two main advantages over other approaches:the system is continually improving itself,perhaps eventually surpassing human experts,and learning allows the system to adapt to changes in the underlying problem.The current implementation has five different learning paradigms(agents)that cooperate to improve problem-solving performance.The agents provide advice to a central controller,called TLC.TLC chooses which suggcstion to adopt and performs the appropriate actions.The agents inspect the results of the TLC's actions and use this feedback to learn,improving their future advice.