| Self-organizing emergent multi-agent systems are a solution to reducing the operational expenditures of complex information technology systems However, emergent behavior is inherently unpredictable, sacrificing the guarantee of efficiency for flexibility and scalability. The addition of self-adaptation through an Efficiency Improvement Advisor (EIA) provides the ability to increase efficiency with a minimal impact on flexibility. The acceptance of EIA-adapted emergent systems is limited by the unreliability of this efficiency improvement. This thesis introduces the Risk-Aware Efficiency Improvement Advisor (RA-EIA), which allows for the assessment and management of risk from proposed adaptations. Monte Carlo Simulation is used to assess and reduce the frequency of emergent misbehavior, and Evolutionary Learning of Event Sequences is used to assess and reduce the severity of the extent of emergent misbehavior. The evaluation of the RA-EIA, for Pickup and Delivery Problems, demonstrates that the advised system can be trusted for independent, reliable, and efficient long-term operation. |
