06 knowledge representation (us)

Upload: arunmohan

Post on 21-Feb-2018

214 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/24/2019 06 Knowledge Representation (Us)

    1/31

    Artificial Intelligence

    Knowledge representation

    Fall 2008

    professor: Luigi Ceccaroni

  • 7/24/2019 06 Knowledge Representation (Us)

    2/31

    2

    Introduction

    Knowledge engineers and system

    analysts need to bring knowledge fort

    and make it e!plicit" #Why?$

    %ey display te implicit knowledge about

    a sub&ect in a form tat programmers can

    encode in algoritms and data structures"

    %o make te idden knowledge accessible

    to computers' knowledge-based

    systemsand object-oriented systems

    are needed" 2

  • 7/24/2019 06 Knowledge Representation (Us)

    3/31

    (

    Introduction

    Knowledge)based and ob&ect)oriented systems

    are built around declarative languages:

    Forms of e!pression closer to uman languages

    *uc systems elp to e!press te knowledge ina form tat bot umans and computers can

    understand"

    %is part of te course is about knowledge)baseanalysis and design:

    %o analy+e knowledge about te real world and map

    it to a computable form

    (

  • 7/24/2019 06 Knowledge Representation (Us)

    4/31

    ,

    Logic' ontology and

    computation Knowledge representation #K-$ is an

    interdisciplinary sub&ect tat applies

    teories and tecni.ues from tree fields:

    Logicpro/ides te formal structureand

    rules of inference"

    Ontologydefines te kinds of things that

    existin te application domain"Computationsupports te applicationstat

    distinguis K- from pure pilosopy"

    ,

  • 7/24/2019 06 Knowledge Representation (Us)

    5/31

    1rinciples of knowledge

    representation Knowledge engineering is te application

    of logic and ontology to te task of

    building computable modelsof some

    domainfor some purpose"

    In 33(' tree e!perts in K-' 4a/is'

    *crobe and *+olo/its' wrote a critical

    re/iew and analysis of te state of te art: Fi/e basic principles about knowledge

    representations#K-s$ and teir role in

    artificial intelligence

  • 7/24/2019 06 Knowledge Representation (Us)

    6/31

    5at is a

    knowledge representation6" A surrogate

    7 Imperfect surrogates mean incorrect inferences areine/itable

    2" A set of ontological commitments

    7 Commitment begins wit te earliest coices7 %e commitments accumulate in layers7 -eminder: a K- is not a data structure

    (" A fragmentary theory of intelligent reasoning

    7 5at is intelligent reasoning67 5ic inferences are sanctioned67 5ic inferences are recommended6

    ," A medium for efficient computation

    " A medium of human expression

  • 7/24/2019 06 Knowledge Representation (Us)

    7/31

    A K- is a surrogate

    4escription of someting else

    Abstract' simplified /iew of a domain

    *ymbolic structure wit formal symbol)manipulating

    rules -ules are based only on te syntactic form of te

    representation

    -e.uires specification of mapping to intendedreferent: an interpretation

    Contains simplifying assumptions and inaccuracies

    *usceptible to supporting incorrect reasoningresults

  • 7/24/2019 06 Knowledge Representation (Us)

    8/31

    A K- is a set of ontological

    commitments 5at to consider in tinking about a

    world: concepts' relations' ob&ects !ample: representing an electric circuit

    9Lumped element model;Components wit connections between tem

    ;Signals flowing instantaneously along te connections

    9lectrodynamics model

    ;Signals propagating at finite speeds;Locations of and distances between components

    ;Components troug wic electromagnetic waves flow

    K- is not about data structures

  • 7/24/2019 06 Knowledge Representation (Us)

    9/31

    A K- is a set of ontological

    commitmentsAn ontological commitment is an agreement

    to use a /ocabulary #i"e"' ask .ueries and

    make assertions$ in a way tat is consistent

    #but not complete$ wit respect to teteory specified by an ontology" 5e build

    agents tat commit to ontologies" 5e

    design ontologies so we can sareknowledge wit and among tese agents"

    %om

  • 7/24/2019 06 Knowledge Representation (Us)

    10/31

    A K- is a fragmentary teory of

    intelligent reasoning It pro/ides different strategies for

    reasoning"

    %ese strategies can be used by umans and

    computers"

    It sanctions a set of inferences"

    95at can we infer from wat we know6

    It recommends a set of inferences"

    95at ougt we to infer from wat we know6

  • 7/24/2019 06 Knowledge Representation (Us)

    11/31

    A K- is a medium for efficient

    computation -easoning in macines is a computational

    process:

    =ot te procedural and te declarati/e

    approaces can be transformed to a

    computable form"

    Computational efficiency is a central

    design goal" !pressi/ity and tractability of reasoning

    are traded off"

  • 7/24/2019 06 Knowledge Representation (Us)

    12/31

    A K- is a medium of uman

    e!pression >ow useful is it as a medium of e!pression6

    >ow general is it6

    >ow precise is it6

    For wat tasks does it pro/ide e!pressi/e ade.uacy6

    >ow useful is it as a medium of communication6

    Can we easily 9talk or tink in te representation

    language6

    5at kinds of tings are easily said in te language6

    5at kinds of tings are so difficult to say in te

    language as to be pragmatically impossible6

  • 7/24/2019 06 Knowledge Representation (Us)

    13/31

    K-s /s" data bases

    =ot 9represent knowledge"

    *tandard data bases do notcontain: dis&unctions #e"g"' 9%e ball is eiter red or blue"$

    .uantifiers #e"g"' 9/ery person as two parents"$

    4ata base scema pro/ide some .uantified information 4educti/e data bases include implications

    4ata base researc concerns: fficient access and management of large distributed data

    bases

    Concurrent updating

    K- researc concerns: !pressi/ity

    ffecti/e reasoning

  • 7/24/2019 06 Knowledge Representation (Us)

    14/31

    ,

    5at is a knowledge base

    #K=$6An informal term for a collection of

    information tat includes an ontology as

    one component"

    =esides an ontology' a K= may contain

    information specified in a declarati/e

    language suc as logic or e!pert)system

    rules" It may also include unstructured or

    unformali+ed information e!pressed in

    natural language or procedural code" ,

  • 7/24/2019 06 Knowledge Representation (Us)

    15/31

    Issues in K- researc

    5at knowledge needs to be represented

    to answer gi/en .uestions6

    >ow is incomplete or noisy information

    represented6

    >ow is .ualitati/e or abstracted knowledge

    represented6

    >ow can knowledge be encoded so tat it

    is reusable6

    >ow are assumptions represented and

    reasoned wit6

  • 7/24/2019 06 Knowledge Representation (Us)

    16/31

    Issues in K- researc

    >ow can knowledge be reformulated for a

    gi/en purpose6

    >ow can effecti/e automatic reasoning be

    done wit large)scale knowledge bases6

    >ow can computer)interpretable

    knowledge be e!tracted from documents6

    >ow can knowledge from multiple sources

    be combined and used6

  • 7/24/2019 06 Knowledge Representation (Us)

    17/31

    Issues in K- researc

    %is is a world were massi/e amounts ofdata and applied matematics replace e/eryoter tool:

    ?ut wit e/ery teory of uman bea/ior' fromlinguistics to sociology" Forget ta!onomy' ontology' and psycology"

    5o knows wy people do wat tey do6

    %e point is tey do it' and we can track andmeasure it wit unprecedented fidelity" 5it enoug data' te numbers speak for

    temsel/es"

    Cris Anderson

  • 7/24/2019 06 Knowledge Representation (Us)

    18/31

    8

    >istorical background

    %e words knowledgeand representationa/e

    pro/oked pilosopical contro/ersies for o/er

    200 years"

    00 ="C": *ocrates claims to know /ery little' ifanyting"

    >e destroyed te self)satisfaction of people wo

    claimed to a/e knowledge of fundamental

    sub&ects like: %rut

    =eauty

    @irtue

    ustice8

  • 7/24/2019 06 Knowledge Representation (Us)

    19/31

    3

    >istorical background

    For is impiety in .uestioning cerised

    beliefs' *ocrates was condemned to deat as

    a corrupter of te morals Atenian yout"

    *ocratesB student 1lato establised te sub&ectof epistemology:

    te study of te nature of knowledge and its

    &ustification

    1latoBs student Aristotle sifted te empasis

    of pilosopy from te nature of knowledge to

    te less contro/ersial but more practical

    problem of representing knowledge" 3

  • 7/24/2019 06 Knowledge Representation (Us)

    20/31

    20

    >istorical background

    AristotleBs work resulted in an encyclopedic

    compilation of te knowledge of is day"

    =ut before e could compile tat

    knowledge' e ad to in/ent te words for

    representing it"

    >e establised te initial terminology and

    defined te scope of logic' pysics'metapysics' biology' psycology'

    linguistics' politics' etics' retoric and

    economics" 20

  • 7/24/2019 06 Knowledge Representation (Us)

    21/31

    2

    >istorical background

    %erms tat Aristotle coined or adopted a/e

    become te core of todayBs international

    tecnical /ocabulary:

    category

    metaphor

    hypothesis

    quantity

    quality

    species

    noun

    """ and ten artificial intelligence arri/ed" 2

  • 7/24/2019 06 Knowledge Representation (Us)

    22/31

    arly istory of K- #0Bs ) D0Bs$

    ?rigins

    1roblem sol/ing work primarily at CE and EI%

    Gatural language understanding

    Eany ad hocformalisms 91rocedural /s" 9declarati/e knowledge

    1rocedural: functions' rules' con/entional

    programming languages

    4eclarati/e: logic' 1rolog

    Go formal semantics

  • 7/24/2019 06 Knowledge Representation (Us)

    23/31

    7 *emantic nets; nstructured node)link graps

    ; Go semantics to support interpretation

    ; Go a!ioms to support reasoning

    ; -eference:

    95atBs in a Link: Foundations for *emantic GetsH 5oods' 5" A"

    In -epresentation and nderstanding: *tudies in Cogniti/e *cienceHedited by 4" =obrow and A" CollinsH Academic 1ressH 3D"

    merging paradigms

    #D0Bs ) 80Bs$

  • 7/24/2019 06 Knowledge Representation (Us)

    24/31

    merging paradigms

    #D0Bs ) 80Bs$7 Frames

    ; *tructured semantic nets

    ; ?b&ect)oriented descriptions

    ; 1rototypes

    ; Class)subclass ta!onomies

    ; -eference:

    9A Framework for -epresenting Knowledge E"Einsky

    Mind DesignH " >augeland' editorH EI% 1ressH 38"

  • 7/24/2019 06 Knowledge Representation (Us)

    25/31

    !ample: Frames: class)

    subclass ta!onomy

  • 7/24/2019 06 Knowledge Representation (Us)

    26/31

    !ample: Frames: Classframe

  • 7/24/2019 06 Knowledge Representation (Us)

    27/31

    !ample: Frames: nstance

    frame

  • 7/24/2019 06 Knowledge Representation (Us)

    28/31

    merging paradigms

    #D0Bs ) 80Bs$7 1roduction rule systems

    ; If)ten inference rules7 If #warning)ligt on$ ten #engine o/ereating$

    7 If #warning)ligt on$ ten ##engine o/ereating$0"3$

    ; *ituation)action rules7 If #warning)ligt on$ ten #turn)off engine$

    ; >ybrid procedural)declarati/e representation; =asis for e!pert systems

  • 7/24/2019 06 Knowledge Representation (Us)

    29/31

    merging paradigms

    #D0Bs ) 80Bs$7 ualitati/e pysics

    ; -epresenting and reasoning:

    7 5it incomplete knowledge

    7 About pysical mecanisms

    ; ualitati/e descriptions

    7 Capture distinctions tat make an important

    .ualitati/e difference and ignores oters

    7 Aggregate /alues tat a/e no .ualitati/e

    difference

  • 7/24/2019 06 Knowledge Representation (Us)

    30/31

    merging paradigms

    #D0Bs ) 80Bs$7 *ymbolic Logic

    ; 1rimarily first)order logic9/erybody lo/es somebody sometime"

    #forall 6p #implies #1erson 6p$ #e!ists #6p2 6t$ #and #1erson 6p2$

    #%ime 6t$

    #Lo/es 6p 6p2 6t$$$$$

    ; -esolution teorem pro/ing

  • 7/24/2019 06 Knowledge Representation (Us)

    31/31

    K- in te 30Bs and 00Bs

    7 4eclarati/e representations; asier to cange

    ; Eulti)use

    ; !tendable by reasoning

    ; Accessible for introspection

    7 Formal semantics; 4efines wat te representation means

    ; *pecifies correct reasoning

    ; Allows comparison of representationsJalgoritms7 K- rooted in te study of logics

    ; temporal' conte!t' modal' default' nonmonotonic"""

    7 -igorous teoretical analysis