부산대학교 도서관

For manageability, large agent societies incorporate features that allow for self-healing in the face of individual failures. Individual features may support or conflict with each other; thus, it may not be optimal for them to act completely autonomously. An adaptive control defense coordinator chooses and mediates such actions to repair a damaged agent society or to protect it against future problems is described. The coordinator attempts to determine the real state of the agent system in the face of multiple diagnoses, then selects a course of action from a collection of potentially conflicting defense actions. In this selection process, it takes into account both any uncertainty about the exact nature of a perceived problem and any high-level policy defining the relative importance of application properties such as completeness, timeliness and security. The coordinator described in this paper is model driven, allowing it to be deployed in a variety of settings. This also allows operators to adapt its behavior to evolving situations by making changes to human-understandable profiles of anticipated stresses or by changing high-level policy describing system goals.