knowledgerepresentationissues.pdf

8/14/2019 KnowledgeRepresentationIssues.pdf

1/17

ARTIFICIAL INTELLIGENCE

Ho Chi Minh City University of TechnologyFaculty of Computer Science and Engineering

Hunh Tn t

Email: [email protected] Page: http://www.cse.hcmut.edu.vn/~htdat/


2/17

Ch4: Knowledge Representation IssuesWhat is KR?

Representation and mapping

Approaches to KR

Issues in KR

Slide 2Faculty of Computer Science and Engineering HCMUT

The Frame problem


3/17

What is KR?R. Davis, H. Schrobe, P. Szolovits (1993):

1. A surrogate

2. A set of ontological commitments3. A fragmentary theory of intelligent reasoning


. me um or e c en compu a on5. A medium of human expressions


4/17

Representation and MappingIn order to solve the complex problems encountered inAI, one needs

Knowledge Mechanisms for manipulating that knowledge

Kinds of entities


Facts: things we want to represent (knowledge level)

Representations of facts: things we can manipulate(symbol level)

Representation mapping


5/17

Representation and Mapping

Facts Internal

Representations

Reasoningprograms

English Englishener tion


EnglishRepresentations

Mappings between Facts and Representations


6/17


Initial

facts

desired real reasoning

forwardrepresentation

Final

facts

backward


Internal

representationsof initial facts

mapping

Internal

representationsof final facts

mapping

operationof program


7/17

Representation and MappingAn example using mathematical logic as therepresentational formalism:

Spot is a dog: dog(Spot) Every dog has a tail: x: dog(x) hastail(x)


Use the deductive mechanisms of logic:

hastail(Spot) Spot has a tail


8/17

Representation and MappingFact-representation mapping is not one-to-one

Every dog has a tail vs. All dogs have tails

Good representation can make a reasoning programtrivial

The Mutilated Checkerboard Problem: Consider a


normal checker board from which two squares, inopposite corners, have been removed. The task is tocover all the remaining squares exactly with

dominoes, each of which covers two squares. Nooverlapping, either of dominoes on top of each otheror of dominoes over the boundary of the mutilatedboard are allowed. Can this task be done?


9/17


No. black squares = 30

No. white squares = 32


(a) (b) (c)


10/17

Approaches to KRFour properties of a good presentation of knowledge in aparticular domain:

Representational adequacy Inferential adequacy

Inferential efficienc


Acquisitional efficiency

No single system that optimizes all of the capabilities for

all kinds of knowledge.=> Multiple techniques for KR exist.


11/17

Approaches to KRSimple relational knowledge

Represent declarative facts as a set of relations used

in database systems Provides very weak inferential capabilities

Ma serve as the in ut to owerful inference en ines



12/17

Approaches to KRInheritable knowledge

Objects are organized into classes and classes are

organized in a generalization hierarchy Inheritance is a powerful form of inference, but not

adequate


Mammal

Person

Owen

Nose

Red Liverpool

isa

instance

has-part

uniformcolor team


13/17

Approaches to KRInferential knowledge

Facts represented in a logical form (e.g. First-Order

Logic: FOL), which facilitates reasoning. An inference engine is required.



14/17

Approaches to KRProcedural knowledge

Representation of how to make it rather than what itis

Procedural knowledge can be represented inprograms in many ways:


Code in some programming language, such as LispMay have inferential efficiency , no inferential

adequacy (difficult to write a program that canreason about another programs behaviour),acquisitional efficiency (b/c of the process ofupdating and debugging large pieces of code)


15/17

Approaches to KR Procedural knowledge as production rules

Distinction between declarative and proceduralknowledge is difficult

If:whi wn r il r nk 2 AND


square(file e, rank 3) is empty ANDsquare(file e, rank 4) is empty

Then:

move pawn from square(file e, rank 2) tosquare(file e, rank 4).


16/17

Issues in KRChoosing the Granularity

High-level facts may not be adequate for inference.Substantial work must be done to reduce theknowledge into primitive form.

Low-level primitives may require a lot of storage.



17/17

HomeworkReading

R. Davis, H. Schrobe, P. Szolovits (1993): What is aknowledge representation?


knowledgerepresentationissues.pdf

Documents