Patricia.Mackiewicz@ISL1.RI.CMU.EDU (05/14/87)
SPECIAL SEMINAR
TOPIC: CYCLOPS: A Computational Model of Exploration & Adaptation
In Design
SPEAKER: Dundee Navinchandra, MIT
WHEN: Thursday, May 14, 1987 at 10:00 am
WHERE: Doherty Hall 3313, CMU
ABSTRACT:
A design system has two basic and essential components: a @b(search)
component and a @b(knowledge) based reasoning component. Designs are
generated by searching the state space of designs, and knowledge is
used to keep the search manageable.
@b(Search Component:) The first part of our research has been in
understanding how designers deal with multiple interacting criteria.
Criteria in design problems can be in the form of constraints, goals or
objectives. It is the job of the designer to produce an artifact that
simultaneously satisfies all the criteria. In the process of achieving
this, the designer has to relax constraints and tradeoff among
objectives. Our system uses pareto-optimality to identify and present
the user with critical tradeoffs in the design problem. The program
also helps the designer @b(explore) the design space by systematically
relaxing constraints and looking for interesting alternatives.
@b(Knowledge-based Component:) The second part of our research is
aimed at developing a technique for recognizing and adapting interesting
designs. This is done through a precedents-based reasoning system.
Precedents are frames that hold knowledge about past design experiences
from within and without the current domain. These experiences are used
to recognize interesting designs. A design is labeled as interesting
if its characteristics cause the reminding of a precedent that was
previously labeled as interesting. Precedents are also used to
@b(adapt) designs that have problems. A technique, called @i(demand
posting) has been developed for solving design problems by reasoning
analogically from the database of precedents.
The above ideas have been implemented in the domain of Landscape
Architecture. The program is called CYCLOPS.