ylfink@water.waterloo.edu (ylfink) (01/29/88)
DEPARTMENT OF COMPUTER SCIENCE
UNIVERSITY OF WATERLOO
SEMINAR ACTIVITIES
ARTIFICIAL INTELLIGENCE SEMINAR/Recruiting Seminar
- Thursday, February 4, 1988
Ms. Kathleen D. Cebulka, from the University of
Delaware, will speak on ``Solving Dynamic-Input
Interpretation Problems''.
TIME: 3:30 PM
ROOM: MC 6082
ABSTRACT
Many problems in artificial intelligence can be
viewed as interpretation problems which have the
common goal of producing a solution that ``explains'' a
given input. The solution usually takes the form
of a set of beliefs called a hypothesis. Although a
number of researchers have developed strategies for
handling the static case where the input is fixed,
there are many problems where the input is received
dynamically in relatively small increments. Usually
the problem solver is interacting with a user who
expects a timely response after every input, so it
can not postpone forming a solution while it waits
for more complete information. As a result, the
problem solver must rely on default reasoning and
hypothesis revision techniques since new evidence may
reveal that the current hypothesis is not the final
answer.
This talk describes a characterization of the
solution of dynamic-input interpretation problems as
a search through a hypothesis space. A domain
independent algorithm, called the Hypothesize-Test-
Revise algorithm, is presented and contrasted with
an algorithm that produces a hypothesis which accounts
for an entire sequence of incremental inputs. An
advantage of the Hypothesize-Test-Revise algorithm is
that it provides an efficient strategy for generating
and revising hypotheses in a dynamic environment.