Presentation is loading. Please wait.

Presentation is loading. Please wait.

Sean Lunsford Brian O’Donnell Rick Kass. Table of Contents  Introduction and Background  Description of the Problem  Proposed Solution  Results 

Similar presentations


Presentation on theme: "Sean Lunsford Brian O’Donnell Rick Kass. Table of Contents  Introduction and Background  Description of the Problem  Proposed Solution  Results "— Presentation transcript:

1 Sean Lunsford Brian O’Donnell Rick Kass

2 Table of Contents  Introduction and Background  Description of the Problem  Proposed Solution  Results  Conclusion, Lessons Learned, Further Study  References

3 Introduction, Background  A new design sought for mobile agent communications  Specifically in domain of communications networks  Inspiration came from behavior of ant colonies, use of chemical markers (pheromones)

4 Introduction, Background cont.  Individual agents (ants) not intelligent, but colony displays collective intelligence  Concept of Stigmergy  Model adapted from ant foraging framework

5 The Problem  Mechanism is needed to monitor inter-node connections, quality of service, faults  Assume no network manager exists  How to create connections between destinations, links in a logical network?

6 The Solution: Stigmergy  Individual agents are unintelligent  Collectively, agents exhibit intelligence through information sharing, following each others’ “footsteps”  Chemical messages are shared and built upon

7 The Solution: Chemical Messages  Agents have the ability to emit, receive “chemical” messages in the environment left by other agents  Agent acts on message based on a Receptor Decision Function  Multiple reactions available  Messages have a duration, reactivity to affect agent response

8 The Solution: Agents  Divided into different classes  Route finding  Connection monitoring  Quality of service monitoring

9 Results  Agents were able to “zero in” on faulty components faster than random search  However, false diagnoses sometimes encountered

10 Conclusion, Lessons Learned  System is shown to be effective  Additional analysis needed  Resolution of false diagnoses (through Reinforcement Learning)  Utility of system  Implementation in large-scale networks  Great potential

11 References  White, Tony and Bernard Pagurek. “Towards Multi-Swarm Problem Solving in Networks.” 1998, Ottowa, Ontario, Canada.


Download ppt "Sean Lunsford Brian O’Donnell Rick Kass. Table of Contents  Introduction and Background  Description of the Problem  Proposed Solution  Results "

Similar presentations


Ads by Google