Monday 4/19/93  ;  WeH 7220  ;  3:00 pm

Lonnie will give a practice talk for his thesis proposal.

_____________________________________________________________________________

Type:     Thesis Proposal
Who:      Lonnie Chrisman
Topic:    Representing and Reasoning about Modeling Limitations
Dates:    22-Apr-93
Time:     2:30 PM
Place:    WeH 4623
			ABSTRACT


   World  models  used  by  A.I. systems for planning or reasoning about the
   world must necessarily be abstractions of reality.    The  world  is  too
   complex  to model in complete detail.  Any given model can always be made
   more faithful to reality by including more detail, but at some point  the
   model  becomes large, unwieldy, and hard to use.  We are always forced to
   omit details even though they may be significant in certain situations.

   This work will examine the following thesis:  * The explicit awareness of
   modeling  limitations  by  a  system  allows world models to be used more
   effectively.

   I am developing  a  new  probabilistic  framework  for  representing  and
   reasoning about the limitations of causal world models.  The framework is
   used to derive a measure of how  well-suited  or  poorly-suited  a  given
   model  is  for  answering  a  given  problem  instance.    A system might
   advantageously utilize this information in several different ways.    For
   example,  it  might be used to detect when the world model at its current
   level of abstraction is insufficient for the task at hand, prompting  the
   system  to switch to an alternative model, to refine the current model to
   more detail, or to defer this particular problem for human assistance.

   During the proposal talk, I will discuss the framework and how it differs
   from   more   standard   probabilistic   approaches   (such  as  Bayesian
   approaches).  For the thesis, I propose to develop this framework to  the
   point where it is useful, demonstrating an implementation in at least two
   application domains.  The first will be a monitoring/diagnosis task based
   on  a  real Space Shuttle missions operations task.  The second will be a
   temporal projection task requiring the coherent reasoning  about  actions
   at different levels of granularity.

Thesis Committee:
                            Tom Mitchell, Co-Chair
                            Reid Simmons, Co-Chair
                                  Matt Mason
                      Glenn Shafer, Princeton University

           A copy of this proposal has been posted in the CS Lounge.