{ Knowledge   Representation Vimal  Atreya  Ramaka 

Ñ  Knowledge  is  the  body  of  facts  and  principles.   Ñ  Knowledge  can  be  language,  concepts,   procedures,  rules,  ideas,  abstractions,  places,   customs,  and  so  on.   Ñ  Study  of  knowledge  is  called  Epistemology. Knowledge 

Ñ  Deals  with  understanding,  designing  and   implementing  ways  of  representing   information. Knowledge  Representation 

Ñ  A  Knowledge  Representation  is  not  a  data   structure Knowledge  Representation 

Ó  A  multidisciplinary  field  applies  theories  and   techniques  from: Ô  Logic Ô  Ontology Ô  Computation Knowledge  Representation 

Ñ  Provides  the  formal  structure  of  and  rules  of   inference Logic 

Ñ  Study  of  correct  reasoning Ñ  Not  a  particular  language Ñ  There  are  many  systems  for  logic Logic 

Ñ  Vocabulary Ó  Collection  of  symbols  represented  as  chars,   words,  icons  or  even  sounds Ñ  Syntax Ó  Rules  that  determine  how  symbols  combine  to   form  well-­‐‑formed  sentences Logic 

Ñ  Semantics Ó  Theory  of  reference  that  determines  how  the   constants  and  variables  relate  to  things  in  the   universe  of  discourse Ñ  Inference Ó  rules  that  determine  how  paOerns  are  generated   from  others Logic 

Ñ  Defines  the  kinds  of  things  that  exist  in  the   application  domain Ontology 

Ñ  Supports  the  applications  that  distinguish   knowledge  representation  from  pure   philosophy Computation 

Ñ  Implementational Ó  Data  structures  such  as  atoms,  pointers,  lists  and   other  programming  notations   Ñ  Logical Ó  symbolic  logic  propositions,  predicates,   variables,  quantifiers  and  Boolean  operations  are   included.   Levels  of  Representation 

Ñ  Epistemological   Ó  defining  concept  types  with  subtypes,   inheritance,  and  structuring  relations   Ñ  Conceptual Ó  defining  concept  types  with  subtypes,   inheritance,  and  structuring  relations  .   Levels  of  Representation 

Ñ  A  knowledge  representation  is  a  surrogate Ó  Symbols  are  used  to  represent  external  things   that  cannot  be  stored  in  a  computer,  i.e.  physical   objects,  events,  and  relationships.  Symbols  are   surrogates  for  the  external  things.  Symbols  and   links  between  them  form  a  model  of  the  external   system  that  can  be  manipulated  to  simulate  it  or   reason  about  it. Principles  of  Representation 

Ñ  A  knowledge  representation  is  a  set  of   ontological  commitments Ó  Ontology  is  the  study  of  existence.  Thus,   ontology  determines  the  categories  of  things  that   exist  or  may  exist  in  an  application  domain.   Those  categories  set  the  ontological   commitments  of  the  application  designer  or   knowledge  engineer. Principles  of  Representation 

Ñ  A  knowledge  representation  is  a  fragmentary   theory  of  intelligent  reasoning Ó  To  support  reasoning  about  modelled  things  in  a   domain,  a  knowledge  representation  must   describe  their  behaviour  and  interactions.  The   description  constitutes  a  theory  of  the   application  domain.  It  can  be  stated,  for  instance,   as  explicit  axioms  or  compiled  into  computable   programs. Principles  of  Representation 

Ñ  A  knowledge  representation  is  a  medium  for   efficient  computation Ó  Besides  representing  knowledge,  an  Artificial   Intelligence  System  must  encode  knowledge  in  a   form  that  can  be  processed  efficiently  by  the   available  computing  equipment.  Therefore,   developments  in  computer  hardware  and   programming  theory  have  a  great  influence  on   knowledge  representation. Principles  of  Representation 

Ñ  A  knowledge  representation  is  a  medium  for   human  expression Ó  A  good  knowledge  representation  language   should  facilitate  communication  between  the   knowledge  engineers  who  manage  knowledge   tools  and  the  domain  experts  who  understand   the  application  domain.  Domain  experts  should   be  able  to  read  and  verify  the  domain  definitions   and  rules  wriOen  by  knowledge  engineers. Principles  of  Representation 

Ñ  Knowledge  Representation Ó  Logic Ó  Ontology Ó  Computation Ñ  Systems  of  Logic Ñ  Levels  of  Representation Ñ  Principles  of  Representation Overview