WPI Worcester Polytechnic Institute

Computer Science Department

Some Knowledge Representation Issues

CS 534 - Artificial Intelligence

Representations should match user's goals/use
What should be represented?
eg. a dog, Spot, Dogs
Events -- time, cause-effect, participants, states, change
How to -- skills, plans, heuristics
knowledge about what you know
how much, how learned, how reliable, how important
how to use what you know
Using knowledge
Acquisition -- addition, integration, contradiction
Removal -- forgetting, removal, negation
accessing what's relevant
indexing, searching, matching
grouping info -- common properties, common use
linking information -- relationships
Reasoning -- how to use, inferences
Formal -- logic, proof theory, truth
Procedural -- mental simulation
What's the 7th letter of the alphabet?
Analogy -- transformational, derivational
Abstraction -- drop properties
Induction -- specific to general (+ Deduction, + Abduction)
Meta-level -- what knowledge to use when
Relationship between knowledge (representation) & use
integrated or separable?
surface (compiled) or deep?
Scope -- what can be represented?
Grain Size -- in what detail are things represented?
Expressive Adequacy -- can you represent everything you need?
Psychological Validity -- does it make a reasonable theory of how humans do it?
Ontology -- what are the concepts to be represented?
Semantic Primitives -- what to use in the language? (has-part, color-is, type-of, ...)
Unique Encoding -- can something be represented in more than one way?
Modularity -- independence of pieces of knowledge?
Is in the eye of the understander
Implicit/Explicit Knowledge -- what knowledge is explicitly expressed, and what is hidden in the representation itself? (eg. meaning assumed/given by the "interpreter" of the KR).
Declarative/Procedural -- more like examined or more like executed?
Inferential Adequacy -- able to make required inferences?
Inferential Efficiency -- ability to control and focus inferences.
Acquisitional Efficiency -- ability to guide acquisition and integration of new knowledge.

dcb@cs.wpi.edu; Fri Oct 25 20:36:12 EDT 1996