Path: utzoo!utgpu!jarvis.csri.toronto.edu!neat.cs.toronto.edu!marina Newsgroups: ont.events From: marina@ai.toronto.edu (Marina Haloulos) Subject: Dr. N.S. Sridharan, Wednesday 22 November 1989: ARTIFICIAL INTELLIGENC Message-ID: <89Nov13.144138est.2899@neat.cs.toronto.edu> Date: 13 Nov 89 19:42:22 GMT Department of Computer Science, University of Toronto (SF = Sandford Fleming Building, 10 King's College Road) ------------------------------------------------------------- ARTIFICIAL INTELLIGENCE SEMINAR SF1013, at 11:00 a.m., Wednesday 22 November 1989 Dr. N.S. Sridharan FMC Corporation "Real-Time Performance for Interactive AI Systems" Achieving real-time performance in a highly interactive and complex knowledge-based systems, such as the Pilot's Associate, poses serious technical challenges. Novel hardware and software approaches must be brought together to yield the desired performance improvements. We review the current stock of best technical innovations along these lines. We also summarize our experience in designing the architecture for the Pilot's Assocaite (DARPA/US Air Force) and Submarine Operational Automation System (DARPA/US Navy). Real-time requirements are often narrowly described in terms of speed alone. There are four important dimensions that are particularly significant to interactive, decision-aiding systems: - speed, - responsiveness, - timeliness, and - graceful adaptation. Speed is measured as the number of tasks executed per unit time. Responsiveness refers to the ability of the systems to take on new tasks quickly. Timeliness characterizes the system's abilities to conform to task priorities. Graceful adaptation is the system's ability to reset priorities based on changing workload. We describe a knowledge-processing architecture we have designed with all four dimensions of real-time performance in mind. This system, built in Common Lisp and Flavors, is an event-driven blackboard system, geared toward asynchronous parallel processing. This talk will describe the performance probes and metrics used to measure responsiveness and timeliness.