This work presents a framework by which a massive multiagent organization can be controlled and modified without resorting to micromanagement and without needing advanced knowledge of potentially complex organizations. In addition to their designated duties, agents in the proposed framework perform some method of determining optimal traits such as configurations, plans, knowledge bases and so forth. Traits follow survival of the fittest rules in which more successful traits overpower less successful ones. Subproblem partitions develop emergently as successful solutions are disseminated to and aggregated by unsuccessful agents. Provisions are provided to allow the administrator to guide the search process by injecting solutions known to work for a particular agent. The performance of the framework is evaluated via comparison to individual state-space search.
Identifer | oai:union.ndltd.org:siu.edu/oai:opensiuc.lib.siu.edu:dissertations-1395 |
Date | 01 August 2011 |
Creators | McLaughlan, Brian Paul |
Publisher | OpenSIUC |
Source Sets | Southern Illinois University Carbondale |
Detected Language | English |
Type | text |
Format | application/pdf |
Source | Dissertations |
Page generated in 0.0024 seconds