Ph.D. Dissertation Defense
Finding a Temporal Frame Comparison Function
for the Metacognitive Loop
Dean Earl Wright III
9:00am Wednesday 26 July 17 August 2011, ITE325 UMBC
The field of Artificial Intelligence has seen steady advances in cognitive systems. However, many of these systems perform poorly when faced with situations outside of their training. And as the real world is dynamic, this brittleness is a major problem in the field today. Adding metacognition to such systems can improve their operation in the face of perturbations found in dynamic environments. The Metacognitive Loop (MCL) works with a host system, monitoring its sensors and expectations. When a failure is indicated, MCL advises the host system on corrective actions.
Differing amounts of metacognition can be made available to the host system. At the lowest level no assistance is given. Above that are rule-based systems with hard-coded responses to stimulations. Next are evaluative systems that weigh multiple inputs using neural network or other techniques. At each level, the metacognition system can provide useful assistance in more situations.
The next level of metacognition adds a temporal dimension. A metacognitive system that has no concept of time has to either treat each new problem as (1) a symptom of previously encountered problems or (2) a completely new problem. Both of these extremes lead to providing poor advice to the host system. Thus, the temporal level of metacognition needs a more discriminating test to compare the current situation encountered by the host system with previous problems.
Several algorithms were developed to find the one that would provide the best performance for a simulated Mars Rover that faced with a dynamic environment with multiple perturbations. Several methods were found that allow MCL to provide good advice most of the time and allow the Mars Rover to successfully complete complex tasks.
Committee:
- Dr. Tim Oates (chair)
- Dr. Marie desJardins
- Dr. Tim Finin
- Dr. Anupam Joshi
- Dr. Don Perlis (UMCP)
