[ CogSci Summaries home | UP | email ]

Bhatta, S. R. & A. K. Goel (1997). Design patterns: A computational theory of analogical design. In the Proceedings of IJCAI-97 workshop on "Using Abstraction and Reformulation in Analogy."

  author = 	 {Sambasiva R. Bhatta and Ashok K. Goel},
  title = 	 {Design Patterns: {A} Computational Theory of
                  Analogical Design},
  booktitle = 	 {Proceedings of IJCAI-97, workshop on Using
abstraction and reformulation in analogy},

Author of the summary: Jim Davies, 1999, jim@jimdavies.org

Cite this paper for:

Examples of GSTs: feedback, turning translational motion into rotational motion.

The input of MBA is a set of structural constraints and functional requirements. Output is a structure that satisfies those constraints.

Design analogues are indexed in memory by both the functions that the stored designs deliver and by the structural constraints they satisfy.

A perfect match if input and output states, and property values and their constraints are identical.

If nothing exact is found, then IDEAL retrieves something close and transforms it. GTMs provide knowledge that allows for encapsulation of relationships between modifications and their causal effects. It then uses qualitative simulation to evaluate the transformed designs.

IDEAL evaluates a candidate design by qualitative simulation.

This is how IDEAL works:

Start with problem constraints.

RETRIEVE m1, a model from memory.
  - retrieve the best model you can based on constraints.
  - if that isn't the solution to the problem then go on.

GET m2, the solution to the problem.
  - try modifications
     - use differences between m1 and abstraction of problem
       constraints to get GTM
     - instantiate
     - compose behaviors
     - evaluate (using qualitative simulation)
  - if that fails then have oracle give correct answer m2.

  - do solution abstraction	
  - save as a GTM what about m2 made it better than m1

In the examples in the text, it tries to solve the first problem unsuccessfully (the amplifier problem). The oracle gives the answer, from this it generalizes the GTM of feedback. Then when it gets the problem of the gyroscope, it solves it as a result of having this new GTM.

Design Patterns

GTMs are represented as BF models. It has 2 kinds of knowledge: 1) about patterns of differences between itself and other known and desired models (for use in difference reduction), and 2) patterns of modifications to the internal behaviors of the known designs that are necessary to reduce differences. That is, a mapping of B to F such that you know how changes to B will affect the functions of the device.

Learning GTMs

Upon presentation of a correct design and SBF model, IDEAL can abstract out and come up with a GTM. It goes through, looking for differences in behaviors, and from that can make an abstraction.

Summary author's notes:

Back to the Cognitive Science Summaries homepage
Cognitive Science Summaries Webmaster:
JimDavies (jim@jimdavies.org)
Last modified: Wed Jun 9 15:27:59 EDT 1999