Gary's Talk

	Weh 4601   Gary Pelton   Feb 15 4.30

A computer agent must be able to accomplish some of its tasks even
when facing a large number of possibly interacting tasks. This
proposal investigates how an agent can start on a task and then set it
aside for later continuation. Setting aside a task in this manner
allows the agent to both complete more tasks and handle more tasks. My
agent uses its experience in doing a task to decide when to set aside
(called suspending) the task, and its knowledge about how to do a task
to remember (called reactivation) the task at the appropriate
times. My agent suspends a task when progress on the task is not
feasible, and reactivates the task either when progress is possible
again or when another task is in conflict with the suspended task.  By
suspending important tasks, my agent can interleave work on other less
important tasks with work on the important tasks, thus completing more
tasks. The agent can handle more tasks, because the suspension process
removes a task from being considered by the agent until its
reactivation, thus making the choice of the next task simpler.

This proposal describes an agent doing everyday tasks and how it
suspends and reactivates these tasks. Everyday tasks provide many
opportunities for testing the agent's ability to appropriately
remember suspended tasks because; the tasks exist over a period of
time, the tasks have many types of suspension opportunities, and
conflicting tasks can arise at any time. This agent will be measured
to see the effect of simultaneous tasks on the agent's computational
complexity as the agent completes thousands of its tasks. These
measurements will be taken as both the number of tasks grow (up to
hundreds), and the number of interactions between those tasks
grow. These measurements will help characterize the applicability of
the research to other agents.