Neural/Expert Based Client Server Architecture for MITE ITS.

摘要

A MITE (multi-node, interactive, task-sharing, expert-instruction) system, in general, is responsible for determining task allocation strategies for members of a team who must work together over a computer network to accomplish an overall objective. The MITE system must adapt to changing skill and performance levels of the team members and continually reallocate the tasks to insure optimal team performance. In order to best allocate tasks among team members, the MITE system should maintain internal models of the member's capacity and knowledge relating to the tasks which that member may be allocated. The system should also provide expert recommendations and instruction for team members to improve performance. In this Phase I study, while investigating the application of hybrid neural network/knowledge base strategies to the problem of MITE systems, we also look for foundation technologies that can be applied to current or future commercial products with high potential returns. Designing the MITE system with an object-oriented clint/server architecture provides the necessary reusability of code objects for a variety of application domains. For example, the complete MITE system can be used for both Army weapons systems, such as Avenger, and large scale manufacturing applications. The task allocation objects can be used within MITE systems or for single user project scheduling. The neural network objects developed for the task allocation module can also be extracted and used for a variety of other optimization problems.