week2 intro knowledge representation

Upload: dominic

Post on 24-Feb-2018

217 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/24/2019 Week2 Intro Knowledge Representation

    1/22

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Lectures on Artificial Intelligence CS364Lectures on Artificial Intelligence CS364

    Introduction toIntroduction to Knowledge RepresentationKnowledge Representation

    13thSeptember 2!

    Dr Bogdan L. rusias!."rusias#surre$.ac.u%

  • 7/24/2019 Week2 Intro Knowledge Representation

    2/22

    13th "o dan #$ %r&sias ' 2

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    ContentsContents

    ( )efining Knowledge Representation

    ( Knowledge Representation Schemes

    ( Semantic *etwor+s

    ( ,a-onom.( /ntolog.

  • 7/24/2019 Week2 Intro Knowledge Representation

    3/22

    13th "o dan #$ %r&sias ' 3

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Knowledge RepresentationKnowledge Representation

    ( 0A representationis a set of conentions abo&t how to

    describe a class of things$ A description ma+es &se of the

    conentions of a representation to describe some partic&lar

    thing$0 inston 12516$

    ( 07ood representations ma+e important ob8ects and relations

    e-plicit9 e-pose nat&ral constraints9 and bring ob8ects and

    relations together0 ibid5 4!

    ( ,he representation principle5

    /nce a problem is described &sing an appropriate representation9

    the problem is almost soled$

  • 7/24/2019 Week2 Intro Knowledge Representation

    4/22

    13th "o dan #$ %r&sias ' 4

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Starting wit& an '(a)pleStarting wit& an '(a)ple

    ( ,he :armer9 ,he :o-9 ,he 7oose and ,he 7rain5

    ,he farmer m&st get a fo-9 a goose and a sac+ of grain across a

    rier9 howeer his boat is small and he can onl. carr. one thing at

    a time$ ;is problem is that if he leaes the fo- with the goose the

    goose will be eaten9 and if he leaes the goose with the grain9 thegrain will be eaten

  • 7/24/2019 Week2 Intro Knowledge Representation

    5/22

    13th "o dan #$ %r&sias ' !

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    7rain

    :o-

    :armer

    7oose

    :armer

    7oose

    7rain

    :o-

    :armer

    :o-7oose

    7rain

    :o-

    7rain

    :armer

    7oose

    :armer

    :o-7rain

    7oose

    7oose

    :armer

    :o-

    7rain

    :armer

    7oose

    :o-

    7rain

    :armer

    7oose

    :o-

    7rain

    :o-

    :armer

    7oose

    7rain

    :armer

    :o-

    7oose

    7rain

  • 7/24/2019 Week2 Intro Knowledge Representation

    6/22

    13th "o dan #$ %r&sias ' 6

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Knowledge Representation Sc&e)esKnowledge Representation Sc&e)es

    ( A n&mber of +nowledge representation schemes or formalisms hae

    been &sed to represent the +nowledge of h&mans in a s.stematic

    manner$ ,his +nowledge is represented in a %nowledge !ases&ch that

    it can be retrieed for soling problems$ Amongst the well?established

    +nowledge representation schemes are5

    =rod&ction R&les

    Semantic *etwor+s

    :rames

    Concept&al )ependenc. 7rammar

    Concept&al 7raphs

    /ntolog.

    =redicate and @odal #ogic

    Concept&al or ,erminological #ogics

    @# B R):

  • 7/24/2019 Week2 Intro Knowledge Representation

    7/2213th "o dan #$ %r&sias '

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Se)antic *etwor%sSe)antic *etwor%s

    ( Ross D&illian 166 and 16E was among the earl. AI

    wor+ers to deelop a comp&tational model which

    represented 0concepts0 as &ierarc&icalnetwor+s$

    ( ,his model was amended with some additional

    ps.chological ass&mptions to characterise the str&ct&re of

    Fh&manG semantic memor.$

  • 7/24/2019 Week2 Intro Knowledge Representation

    8/2213th "o dan #$ %r&sias ' E

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Se)antic *etwor%sSe)antic *etwor%s

    ( Collins and D&illian 16 proposed that5

    Concepts can be represented as hierarchies of inter?

    connected concept nodes e$g$ animal9 bird9 canar.

    An. concept has a n&mber of associated attrib&tes at a gien leel

    e$g$ animal ??H has s+in eats etc$

    Some concept nodes are superordinatesof other nodes e$g$animal H bird and some are su!ordinatescanar. J bird

  • 7/24/2019 Week2 Intro Knowledge Representation

    9/2213th "o dan #$ %r&sias '

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Se)antic *etwor%sSe)antic *etwor%s

    ( :or reasons of cogniti"e econo)$9 s&bordinates in&erit

    all the attrib&tes of their s&perordinate concepts$

    ( Some instances of a concept are e(ceptedfrom the

    attrib&tes that help Fh&mansG to define the s&perordinates

    e$g$ ostrich is e-cepted from fl.ing

    ( %ario&s Fps.chologicalG processes searc&these hierarchies

    for information abo&t the concepts represented$

  • 7/24/2019 Week2 Intro Knowledge Representation

    10/2213th "o dan #$ %r&sias ' 1

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Se)antic *etwor% '(a)pleSe)antic *etwor% '(a)ple

    birdcan fl.9 &as wings9

    &as feathers

    salmonlayseggsswims&pstream9

    ispin+9 isedible

    ostrichrunsfast9 cannotfly9

    istall

    canar.can sing9 is.ellow

    fishcan swim9 &as fins9 &as gills

    ani)alcanbreathe9 caneat9

    &as s+in

    is+a

    is+a

    is+a

    is+a

    is+a

  • 7/24/2019 Week2 Intro Knowledge Representation

    11/2213th "o dan #$ %r&sias ' 11

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Se)antic *etwor% '(a)pleSe)antic *etwor% '(a)ple

    ( :rom the aboe ta-onomic organisation of +nowledge abo&t a n&mber

    of different animals9 one can concl&de9 b. 0in&eriting properties

    down t&e ta(ono)$09 that canaries9 ostriches and salmon all hae s+in

    and can breathe$

    ( "&t we as h&mans can also ma+e e(ceptionsto inherited properties in

    that we can represent an not?fl.ing bird in a s&b? hierarch. of birds

    b. simpl. noting the e-ception9 can't fly$

    ( Collins and D&illian carried o&t a n&mber of tests on h&man s&b8ectsand fo&nd that the s&b8ects recognise propositions lower down the

    hierarch. canar. is a .ellow bird more readil. than propositions

    higher &p the hierarch. canar. has s+in$

  • 7/24/2019 Week2 Intro Knowledge Representation

    12/2213th "o dan #$ %r&sias ' 12

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Defining In&eritanceDefining In&eritance

    ( AI researchers hae refined the notion of inheritance5

    It is called a specialised inferencing techni&e 0for representing

    properties of classes9 e-ceptions to inherited properties9 m&ltiple

    s&perclasses9 and str&ct&red concepts with specific relationsamong the str&ct&ral elements0 ,o&retL+. 1256$

  • 7/24/2019 Week2 Intro Knowledge Representation

    13/2213th "o dan #$ %r&sias ' 13

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Se)antic *etwor%sSe)antic *etwor%s

    ( A semantic networ+ is a str&ct&re for representing +nowledge as a

    pattern of interconnected nodesand arcs$ *odes in the net represent

    conceptsof entities9 attrib&tes9 eents9 al&es$ Arcs in the networ+

    represent relations&ipsthat hold between the concepts$

    ( A semantic networ+ is a graph theoretic data str&ct&re whose nodes

    represent word senses and whose arcs e-press se)antic relations&ips

    between these word senses$

    ( D&illian gae an acco&nt9 perhaps first &sed b. a comp&ter scientist9 ofthe associate feat&res of h&man memor. that incorporated a spreading

    actiation model of comp&tation$

  • 7/24/2019 Week2 Intro Knowledge Representation

    14/2213th "o dan #$ %r&sias ' 14

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    *etwor%s and ,-eaning, Representation*etwor%s and ,-eaning, Representation

    ( ,he bios.stematic notions of ta(ono)ies9 where the

    concept of s&perordinates and s&bordinates pla.s a ma8or

    role on the +nowledge representation literat&re$

    ( ,A/*/@M /: #I:N5 ,he ta-onomic organisation of

    species in a hierarchical str&ct&re5 Kingdom H =h.l&m diision in botan. H Class H /rder H :amil.

    H 7en&s H Species

    ( Carol&s #innea&s c$1Eth cent&r. Swedish botanist

    deised the s.stem of binomial nomenclat&re &sed fornaming species5

    each species has a two?part #atin name9 formed b. appending a

    specific epithet to the gen&s name$

    ,he latter is capitalised and both parts italicised$

  • 7/24/2019 Week2 Intro Knowledge Representation

    15/2213th "o dan #$ %r&sias ' 1!

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    D/ S0/AR-A1L'

    BR'AD-0LD

    I*2'S2I*ALBAC2'RI0-

    1*D AL/A'

    KI*/D- Animaliaanimals

    =lantae plants :&ngi f&ngi =ro+ar.otaebacteria

    =rotoctista algae9protoLoa9 slimmo&lds

    1L0- Chordata @agnolioph.ta O.gom.cota /mnibacteria Chloroph.ta

    CLASS @ammalia Rosidae O.gom.cetes Nnterobacteria N&con8&gatae

    RD'R Carniora Sapindales @acorales N&bacteriales O.gnematales

    5A-IL Canidae Aceraceae @&coraceae E. colidoes not&a"e a fa)il$

    classification7

    O.gnemataceae

    /'*0S Canis Acer Rhizopus Escherichia Spirogyra

    S1'CI'S C. familiaris A.saccharum R. stolonifer E. Coli S. crassa

    -odern 2a(ono)$ '(a)ple-odern 2a(ono)$ '(a)ple

  • 7/24/2019 Week2 Intro Knowledge Representation

    16/2213th "o dan #$ %r&sias ' 16

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    Representation and Bios$ste)aticsRepresentation and Bios$ste)atics

    ( or+ in +nowledge representation has been infl&enced b.

    +e. notions in bios.stematics$

    ( ;oweer9 there are cr&cial differences between what a

    ta-onomist does and a +nowledge engineer does$

    ( ,he +e. difference is that of the intended a&dience in the

    two cases5

    for the ta-onomist the a&dience is intelligent and h&man

    and for the +nowledge engineer the primar. 0a&dience0 is a

    comp&ter s.stem9 or more acc&ratel. the representation program$

    A ifi i l lli CS364

  • 7/24/2019 Week2 Intro Knowledge Representation

    17/2213th "o dan #$ %r&sias ' 1

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    ntolog$ntolog$

    ( )efinition5

    P,he science or st&d. of being that department of metaph.sics

    which relates to the being or essence of things9 or to being in the

    abstract$Q /N) online9 http5BBwww$oed$comB

    ifi i l lliA ifi i l I lli CS364

  • 7/24/2019 Week2 Intro Knowledge Representation

    18/2213th "o dan #$ %r&sias ' 1E

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    ntolog$ntolog$

    ( AI e-perts9 li+e ,om 7r&ber9 s&ggest that5

    0In the conte-t of +nowledge sharing9 I &se the term ontolog. to

    mean aspecification of a conceptualization$ ,hat is9 an ontolog. is

    a description li+e a formal specification of a program of the

    concepts and relationships that can e-ist for an agent or a

    comm&nit. of agents$ ,his definition is consistent with the &sage

    of ontolog. as set?of?concept?definitions9 b&t more general$ And it

    is certainl. a different sense of the word than its &se in

    philosoph.$0 Cited from www?+sl$stanford$ed&B+stBwhat?is?an?

    ontolog.$html site isited 12BB!

    A ifi i l I lli CS364A ifi i l I lli CS364

  • 7/24/2019 Week2 Intro Knowledge Representation

    19/2213th "o dan #$ %r&sias ' 1

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    ntolog$ntolog$

    ( Also ,om 7r&ber9 s&ggest that5

    0An ontolog. is an e-plicit specification of a concept&aliLation$ ,he term

    is borrowed from philosoph.9 where an /ntolog. is a s.stematic acco&nt

    of N-istence$ :or AI s.stems9 what e-ists is that which can be

    represented$ hen the +nowledge of a domain is represented in a

    declaratie formalism9 the set of ob8ects that can be represented is calledthe uni"erse of discourse$ ,his set of ob8ects9 and the describable

    relationships among them9 are reflected in the representational ocab&lar.

    with which a +nowledge?based program represents +nowledge$ ,h&s9 in

    the conte-t of AI9 we can describe the ontolog. of a program b. defining

    a set of representational terms$ In s&ch an ontolog.9 definitions associate

    the names of entities in the &nierse of disco&rse e$g$9 classes9 relations9f&nctions9 or other ob8ects with h&man?readable te-t describing what the

    names mean9 and formal a-ioms that constrain the interpretation and well?

    formed &se of these terms$ :ormall.9 an ontolog. is the statement of a

    logical theor.$0 Cited from www?+sl$stanford$ed&B+stBwhat?is?an?

    ontolog.$html site isited 12BB!

    A ifi i l I lli CS364A tifi i l I t lli CS364

  • 7/24/2019 Week2 Intro Knowledge Representation

    20/2213th "o dan #$ %r&sias ' 2

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    ntolog$ as a Specification -ec&anis)ntolog$ as a Specification -ec&anis)

    ( A bod. of formall. represented +nowledge is based on a

    conceptuali8ation5 the ob8ects9 concepts9 and other entities

    that are ass&med to e-ist in some area of interest and the

    relationships that hold among them 7enesereth T

    *ilsson9 1E$( A concept&aliLation is an abstract9 simplified iew of the

    world that we wish to represent for some p&rpose$

    ( Ner. +nowledge base9 +nowledge?based s.stem9 or

    +nowledge?leel agent is committed to someconcept&aliLation9 e-plicitl. or implicitl.$

    A tifi i l I t lli CS364A tifi i l I t lli CS364

  • 7/24/2019 Week2 Intro Knowledge Representation

    21/2213th "o dan #$ %r&sias ' 21

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    ntolog$ntolog$

    A tifi i l I t lli CS364A tifi i l I t lli CS364

  • 7/24/2019 Week2 Intro Knowledge Representation

    22/2213th " d # % i ' 22

    Artificial Intelligence CS364Artificial Intelligence CS364

    Knowledge RepresentationKnowledge Representation

    ClosingClosing

    ( D&estionsUUU

    ( Remar+sUUU

    ( Comments>>>( Nal&ation>