perm filename CHAP4[DIS,DBL] blob
sn#207266 filedate 1976-03-19 generic text, type C, neo UTF8
COMMENT ā VALID 00005 PAGES
C REC PAGE DESCRIPTION
C00001 00001
C00002 00002 .NSECP(AM's Knowledge: Concepts and Heuristics)
C00004 00003
C00008 00004 .SSEC(Representation of Concepts)
C00009 00005 .SSEC(Heuristic Guidance)
C00010 ENDMK
Cā;
.NSECP(AM's Knowledge: Concepts and Heuristics)
.ONCE TURN ON "{}"
This chapter contains more than you want to know about AM's innards.
After a brief overview, we'll look at the way concepts are
represented {SECNUM}.2 and the whole business about "heuristic rules".
Section {SECNUM}.4 presents a condensed version of the knowledge AM starts
with.
Chapter 5 gives several examples illustrating not only AM's control structure,
butalso the changes in its state of knowledge at various moments.
The theoretical issues discussed in chapters 3-5
will probably raise as many questions as they
answer (e.g., "how can one judge the performance of a program like
AM?"). The whole of Chapter 6 will be spent answering some of these.
.SSEC(Representation of Concepts)
How shall a concept be represented internally?
.B
Every concept consists merely of a bundle of facets.
Describe the idea of a fixed universal set of facets.
Illustrate the particular set of facets with a typical concept (composition)
Some particular points not to omit:
The interestingness facet; the worth facet.
The linked nature of the generalizations/specializations facets.
The linked nature of the examples/up-isa facets.
Equations which show how to efficiently locate desirable things:
EXS(z) = ((z . SPEC*) . EXS) . SPEC*
.E
.SSEC(Heuristic Guidance)
What about the heuristics?
.B
Elicited in the context of a specific concept, so keep them tacked onto
that concept, as if they were normal facets.
Assumes a linearity which may not really be there! Alternative is to
maintain 2ān bundles of heuristics, and this is untenable.
Represented as situation-action rules. Describe left/right sides; exs.
.E