Web Intelligence (WI) Web Intelligence (WI) Some Research - - PDF document

▶
web intelligence wi web intelligence wi
SMART_READER_LITE
LIVE PREVIEW

Web Intelligence (WI) Web Intelligence (WI) Some Research - - PDF document

Nov 28, 2023 152 likes •441 views

Web Intelligence (WI) Web Intelligence (WI) Some Research Challenges Some Research Challenges [IJ [IJ CAI CAI03 Invited Talk] 03 Invited Talk] Jiming Liu Liu Jiming Web Intelligence Consortium (WIC) Web Intelligence Consortium (WIC)

slide-1
SLIDE 1

Web Intelligence (WI) Web Intelligence (WI)

Some Research Challenges Some Research Challenges

[IJ CAI [IJ CAI’03 Invited Talk] 03 Invited Talk]

Jiming Jiming Liu Liu

Web Intelligence Consortium (WIC) Web Intelligence Consortium (WIC) & Department of Computer Science Department of Computer Science Hong Kong Baptist University Hong Kong Baptist University

j i m i ng j i m i ng@ c om p. @ c om

  • p. hkbu
  • hkbu. edu
  • edu. hk

hk

I JCAI I JCAI ’ 03 8/ 15/ 2003 03 8/ 15/ 2003

2003-8-29 2

Acknowledgements and Credits Acknowledgements and Credits

  • Profs.
  • Profs. Ning Zhong

Ning Zhong, , Yiyu Yao Yiyu Yao, , Edward A. Edward A. Feigenbaum Feigenbaum, , Setsuo Ohsuga Setsuo Ohsuga, , Benjamin Benjamin Wah Wah, Philip Yu, , Philip Yu, Lotfi Lotfi A.

  • A. Zadeh

Zadeh, and , and Xindong Xindong Wu, etc. Wu, etc.

  • WIC Technical Committee

WIC Technical Committee

  • WIC Research Centers in Australia, Canada, India

WIC Research Centers in Australia, Canada, India Japan, and Spain, among others Japan, and Spain, among others

  • Students and Post

Students and Post-doc/Visitors at doc/Visitors at

  • A.M.D. Lab (HKBU)

A.M.D. Lab (HKBU)

  • AAMAS/AOC Group (HKBU)

AAMAS/AOC Group (HKBU)

slide-2
SLIDE 2

2003-8-29 3

  • The preparation has benefited from a peaceful

The preparation has benefited from a peaceful break in break in Grand Canyon, Grand Canyon, on the way to IJCAI

  • n the way to IJCAI’03

03 …

2003-8-29 4

Outline Outline

Background

Background

Challenges

Challenges

Semantic Web + Planning

Semantic Web + Planning

Distributed Agents + Coordination

Distributed Agents + Coordination

Social Networks + Self

Social Networks + Self-Organization Organization

Issues and Directions

Issues and Directions

slide-3
SLIDE 3

2003-8-29 6

What is Web Intelligence (WI)? What is Web Intelligence (WI)?

WI WI explores the fundamental roles as well explores the fundamental roles as well as practical impacts of as practical impacts of

  • Artificial Intelligence (AI)

Artificial Intelligence (AI) (e.g., knowledge representation, (e.g., knowledge representation, planning, knowledge discovery, agents, planning, knowledge discovery, agents, and social intelligence and social intelligence) and and

  • Advanced Information Technology (IT)

Advanced Information Technology (IT) (e.g., wireless networks, ubiquitous e.g., wireless networks, ubiquitous devices, social networks, and devices, social networks, and data/knowledge grids) data/knowledge grids)

  • n the next generation of Web
  • n the next generation of Web-empowered

empowered systems, environments, and activities systems, environments, and activities

Zhong Zhong, N , Liu, J , and , N , Liu, J , and Yao Yao, Y Y (eds.) , Y Y (eds.) Web Intelligence Web Intelligence, Springer, 2003 , Springer, 2003

slide-4
SLIDE 4

2003-8-29 7 2003-8-29 8

Four Levels of WI Support Four Levels of WI Support

Internet-level communication, infrastructure, and security protocols

Level-1

Interface-level multi-media presentation standards

Level-2

Knowledge-level information processing and management tools

Level-3

Application-level ubiquitous computing and social intelligence utilities

Level-4 support

slide-5
SLIDE 5

WI Challenge #1 WI Challenge #1

Semantic Web + Planning Semantic Web + Planning

2003-8-29 10

Semantic Web Semantic Web

  • Ontology:

Ontology: Define Define meanings and relationships of meanings and relationships of vocabularies vocabularies (in terms of classes and properties) (in terms of classes and properties)

  • Semantic Web:

Semantic Web: Add semantic meanings to Web information Add semantic meanings to Web information based on pre based on pre-defined ontology defined ontology

  • Benefits:

Benefits: Enable better human Enable better human-computer communications as computer communications as well as software agents access well as software agents access

  • Example:

Example: DARPA Agent Markup Language (DAML) DARPA Agent Markup Language (DAML) SGML XML HTML RDF SHOE OIL

Content Definition Content Representation

Web DAML

Ontology/Agent Markup DAML+OIL

OWL

slide-6
SLIDE 6

2003-8-29 11

<?xm l ve r s i on="1. 0"?> <r df : RDF xm l ns: rdf ="ht t p: / / www. w3. or g/ 1999/ 02/ 22 - r df - s ynt ax- ns #" xm l ns: r df s ="ht t p: / / www. w3. or g/ 2000/ 01/ r df - s c he m a#" xm l ns: dam l ="ht t p: / / www. dam l . or g/ 2001/ 03/ dam l +oi l #" xm l : bas e ="ht t p: / / nt s e r ve r . hom e i p. ne t / Honor Pr oj e c t / DAM L/ l ogi c - ont o#"> <dam l : Ont ol ogy dam l : ve r s i onI nf o="1. 0"> <dam l : c om m e nt >An e xam pl e f or l ogi c ont ol ogy</ dam l : c om m e nt > </ dam l : Ont ol ogy> <! - - Te r m - - > <r df s: Cl as s rdf : I D="Te r m"/ > <! - - Var i abl e - - > <r df s: Cl as s rdf : I D="Var i abl e "> <dam l : s ubCl as s Of r df : r e s our c e ="#Te r m "/ > </ r df s : Cl as s > <! - - Cons t ant - - > <r df s: Cl as s rdf : I D="Cons t ant "> <dam l : s ubCl as s Of r df : r e s our c e ="#Te r m "/ > </ r df s : Cl as s > <dam l : Dat at ype Pr ope r t y r df : I D="val ue "> <dam l : dom ai n rdf : r e s our c e ="#Cons t ant "/ > <dam l : r ange r df : r e s our c e ="ht t p: / / www. w3. or g/ 2000/ 10/ XM LSc he m a #s t r i ng"/ > </ dam l : Dat at ype Pr ope r t y> <! - - Pr e di c at e - - > <r df s: Cl as s rdf : I D="Pr e di c at e "> <dam l : s ubCl as s Of r df : r e s our c e ="#Te r m "/ > </ r df s : Cl as s > <dam l : O bj e c t Pr ope r t y r df : I D="ar gum e nt s "> <dam l : dom ai n rdf : r e s our c e ="#Pr e di c at e "/ > <dam l : r ange r df : r e s our c e ="#Te r m "/ > </ dam l : O bj e c t Pr ope r t y > <! - - Com m

  • n Pr ope r t y - - >

<dam l : Dat at ype Pr ope r t y r df : I D="nam e "> <dam l : dom ai n rdf : r e s our c e ="#Var i abl e "/ > <dam l : dom ai n rdf : r e s our c e ="#Pr e di c at e "/ > <dam l : r ange r df : r e s our c e ="ht t p: / / www. w3. or g/ 2000/ 10/ XM LSc he m a #s t r i ng"/ > </ dam l : Dat at ype Pr ope r t y> </ rdf : RDF>

2003-8-29 12

Planning Planning

Planning example:

Planning example: STRIPS STRIPS

States:

States: conjunctions of ground literals conjunctions of ground literals

At ( H

  • m

e ) ^ Se l l ( Supe r m ar ke t , Banana) ^ Se l l ( Supe r m ar ke t , M i l k) ^ Se l l ( H ar dwar e St or e , Dr i l l ) H ave ( Dr i l l ) ^ H ave ( M i l k) ^ H ave ( Banana) ^ At ( H

  • m

e ) At ( ?x) ^ Se l l s ( ?x, M i l k)

Goals:

  • als: conjunctions of literals

conjunctions of literals (possibly

(possibly containing variables) containing variables)

slide-7
SLIDE 7

2003-8-29 13

Operators in STRIPS Operators in STRIPS

Ac t i on: Buy( ?x) Pr e c ondi t i ons : At ( ?s t or e ) ^ Se l l ( ?s t or e , ?x) Ef f e c t s : addi t i ons : H ave ( ?x) de l e t i ons : Buy( ?x) At ( ?s t or e ) , Se l l ( ?s t or e , ?x) + H ave ( ?x)

Ac t i on: G

  • ( ?t he r e )

Pr e c ondi t i ons : At ( ?he r e ) Ef f e c t s : addi t i ons : At ( ?t he r e ) de l e t i ons : At ( ?he r e ) G

  • ( ?t he r e )

At ( ?he r e ) + At ( ?t he r e )

  • At ( ?he r e )

2003-8-29 14

POP Algorithm POP Algorithm

Ordering constraint

Ordering constraint

  • Step

Step Si occurs

  • ccurs

before step before step Sj

Casual link

Casual link

  • Si achieves the

achieves the precondition precondition c of

  • f Sj

Open condition

Open condition

  • Precondition that is

Precondition that is not causally linked not causally linked Si Sj

c

Si < Sj

Source: Russell & Norvig

slide-8
SLIDE 8

2003-8-29 15

Semantic Web + Planning Semantic Web + Planning

(with Kelvin Tsang) (with Kelvin Tsang)

  • The planning agent is

The planning agent is goal goal-directed directed

  • A

A plan plan is a sequence of is a sequence of actions actions to achieve to achieve the the goals, given an initial state goals, given an initial state

  • A

A logic logic-based language based language is used to describe the is used to describe the problem problem

  • General planner is based on

General planner is based on Partial Order Planning Partial Order Planning (POP), coupled with (POP), coupled with heuristic search heuristic search

  • Meanings and relationships of the words in the

Meanings and relationships of the words in the documents are specified in documents are specified in ontologies

  • ntologies
  • Planning information

Planning information is interpreted from semantic is interpreted from semantic Web documents Web documents

2003-8-29 16

<?xm l ve r s i on="1. 0"?> <r df : RDF xm l ns: rdf ="ht t p: / / www. w3. or g/ 1999/ 02/ 22 - r df - s ynt ax- ns #" xm l ns: r df s ="ht t p: / / www. w3. or g/ 2000/ 01/ r df - s c he m a#" xm l ns: dam l ="ht t p: / / www. dam l . or g/ 2001/ 03/ dam l +oi l #" xm l : bas e ="ht t p: / / nt s e r ve r . hom e i p. ne t / Honor Pr oj e c t / DAM L/ l ogi c - ont o#"> <dam l : Ont ol ogy dam l : ve r s i onI nf o="1. 0"> <dam l : c om m e nt >An e xam pl e f or l ogi c ont ol ogy</ dam l : c om m e nt > </ dam l : Ont ol ogy> <! - - Te r m - - > <r df s: Cl as s rdf : I D="Te r m"/ > <! - - Var i abl e - - > <r df s: Cl as s rdf : I D="Var i abl e "> <dam l : s ubCl as s Of r df : r e s our c e ="#Te r m "/ > </ r df s : Cl as s > <! - - Cons t ant - - > <r df s: Cl as s rdf : I D="Cons t ant "> <dam l : s ubCl as s Of r df : r e s our c e ="#Te r m "/ > </ r df s : Cl as s > <dam l : Dat at ype Pr ope r t y r df : I D="val ue "> <dam l : dom ai n rdf : r e s our c e ="#Cons t ant "/ > <dam l : r ange r df : r e s our c e ="ht t p: / / www. w3. or g/ 2000/ 10/ XM LSc he m a #s t r i ng"/ > </ dam l : Dat at ype Pr ope r t y> <! - - Pr e di c at e - - > <r df s: Cl as s rdf : I D="Pr e di c at e "> <dam l : s ubCl as s Of r df : r e s our c e ="#Te r m "/ > </ r df s : Cl as s > <dam l : O bj e c t Pr ope r t y r df : I D="ar gum e nt s "> <dam l : dom ai n rdf : r e s our c e ="#Pr e di c at e "/ > <dam l : r ange r df : r e s our c e ="#Te r m "/ > </ dam l : O bj e c t Pr ope r t y > <! - - Com m

  • n Pr ope r t y - - >

<dam l : Dat at ype Pr ope r t y r df : I D="nam e "> <dam l : dom ai n rdf : r e s our c e ="#Var i abl e "/ > <dam l : dom ai n rdf : r e s our c e ="#Pr e di c at e "/ > <dam l : r ange r df : r e s our c e ="ht t p: / / www. w3. or g/ 2000/ 10/ XM LSc he m a #s t r i ng"/ > </ dam l : Dat at ype Pr ope r t y> </ rdf : RDF>

slide-9
SLIDE 9

2003-8-29 17

Ontology Modeling Ontology Modeling

<! - - Pl anni ng dom ai n - - > <rdf s : C l as s rdf : I D ="Dom ai n"/ > <dam l : O bj ect Propert y r df : I D="ope r at or s"> <dam l : dom ai n rdf : res ource="#Dom ai n"/ > <dam l : range rdf : r e s our c e ="#O perat or "/ > </ dam l : O bj ect Propert y> <dam l : O bj ect Propert y r df : I D="i ni t i al "> <dam l : dom ai n rdf : res ource="#Dom ai n"/ > <dam l : range rdf : r e s our c e ="l ogi c: Predi cat e"/ > </ dam l : O bj ect Propert y> <dam l : O bj ect Propert y r df : I D="goal "> <dam l : dom ai n rdf : res ource="#Dom ai n"/ > <dam l : range rdf : r e s our c e ="l ogi c: Predi cat e"/ > </ dam l : O bj ect Propert y>

Domain

Operator

  • perators

Predicate

initial

Predicate

g

  • a

l

2003-8-29 18

Instance Files Instance Files

  • DAML instance files:

DAML instance files: To encode planning To encode planning information in Semantic Web documents that use information in Semantic Web documents that use the vocabularies in the vocabularies in

  • logic ontology (logic

logic ontology (logic-onto.

  • nto.daml

daml)

  • planning ontology (plan

planning ontology (plan-onto.

  • nto.daml

daml)

shopping

shopping-problem. problem.daml daml

Ac t i on: Buy( ?x) Pr e c ondi t i ons : At ( ?s t or e ) ^ Se l l ( ?s t or e , ?x) Ef f e c t s : addi t i ons : H ave ( ?x) de l e t i ons :

slide-10
SLIDE 10

2003-8-29 19

Examples of Operators Examples of Operators

<! <! - -

  • -

Buy( ?x) Buy( ?x) - -

  • - >

<pl an: Ope r at or > <pl an: Ope r at or > <r df s r df s: c om m e nt >Buy t he ?t he r e </ : c om m e nt >Buy t he ?t he r e </ r df s r df s : c om m e nt > : c om m e nt > <pl an: ac t i on> <pl an: ac t i on> <l ogi c : Pr e di c at e l ogi c : Pr e di c at e > <l ogi c : nam e > <l ogi c : nam e > <r xs d r xs d: s t r i ng : s t r i ng r df r df : val ue =" : val ue =" Buy Buy "/ > "/ > </ l ogi c : nam e > </ l ogi c : nam e > <l ogi c : ar gum e nt s <l ogi c : ar gum e nt s rdf rdf : par s e Type par s e Type =" ="dam l dam l : c ol l e c t i on"> : c ol l e c t i on"> <l ogi c : Var i abl e > <l ogi c : Var i abl e > <l ogi c : nam e > <l ogi c : nam e > <r xs d r xs d : s t r i ng : s t r i ng r df r df : val ue =" : val ue ="?x ?x "/ > "/ > </ l ogi c : nam e > </ l ogi c : nam e > </ l ogi c : Var i abl e > </ l ogi c : Var i abl e > </ l ogi c : ar gum e nt s > </ l ogi c : ar gum e nt s > </ </ l ogi c : Pr e di c at e l ogi c : Pr e di c at e > </ pl an: ac t i on> </ pl an: ac t i on> <! <! - -

  • -

Pr e c ondi t i ons Pr e c ondi t i ons - -

  • - >

<pl an: pr e c ondi t i ons pl an: pr e c ondi t i ons r df r df : par s e Type par s e Type =" ="dam l dam l : c ol l e c t i on"> : c ol l e c t i on"> <! <! - -

  • -

At ( ?s t or e ) At ( ?s t or e ) - -

  • - >

<l ogi c : Pr e di c at e l ogi c : Pr e di c at e > <l ogi c : nam e > <l ogi c : nam e > <r xs d r xs d: s t r i ng : s t r i ng r df r df : val ue =" : val ue =" At At "/ > "/ > </ l ogi c : nam e > </ l ogi c : nam e >

Buy( ?x)

2003-8-29 20

<l ogi c : ar gum e nt s <l ogi c : ar gum e nt s rdf rdf : par s e Type par s e Type =" ="dam l dam l : c ol l e c t i on"> : c ol l e c t i on"> <l ogi c : Var i abl e > <l ogi c : Var i abl e > <l ogi c : nam e > <l ogi c : nam e > <r xs d r xs d : s t r i ng : s t r i ng r df r df : val ue =" : val ue ="?s t or e ?s t or e "/ > "/ > </ l ogi c : nam e > </ l ogi c : nam e > </ l ogi c : Var i abl e > </ l ogi c : Var i abl e > </ l ogi c : ar gum e nt s > </ l ogi c : ar gum e nt s > </ </ l ogi c : Pr e di c at e l ogi c : Pr e di c at e > <! <! - -

  • -

Se l l ( ?s t or e , ?x) Se l l ( ?s t or e , ?x) - -

  • - >

<l ogi c : Pr e di c at e l ogi c : Pr e di c at e > <l ogi c : nam e > <l ogi c : nam e > <r xs d r xs d: s t r i ng : s t r i ng r df r df : val ue =" : val ue =" Se l l Se l l "/ > "/ > </ l ogi c : nam e > </ l ogi c : nam e > <l ogi c : ar gum e nt s l ogi c : ar gum e nt s rdf rdf : par s e Type par s e Type =" ="dam l dam l : c ol l e c t i on"> : c ol l e c t i on"> <l ogi c : Var i abl e l ogi c : Var i abl e > <l ogi c : nam e > <l ogi c : nam e > <r xs d r xs d : s t r i ng : s t r i ng r df r df : val ue =" : val ue ="?s t or e ?s t or e "/ > "/ > </ l ogi c : nam e > </ l ogi c : nam e > </ </ l ogi c : Var i abl e l ogi c : Var i abl e > <l ogi c : Var i abl e l ogi c : Var i abl e > <l ogi c : nam e > <l ogi c : nam e > <r xs d r xs d : s t r i ng : s t r i ng r df r df : val ue =" : val ue ="?x ?x "/ > "/ > </ l ogi c : nam e > </ l ogi c : nam e > </ </ l ogi c : Var i abl e l ogi c : Var i abl e > </ </ l ogi c : ar gum e nt s l ogi c : ar gum e nt s > </ </ l ogi c : Pr e di c at e l ogi c : Pr e di c at e > </ </ pl an: pr e c ondi t i ons pl an: pr e c ondi t i ons>

At ( ?s t or e ) Se l l ( ?s t or e , ?x)

slide-11
SLIDE 11

2003-8-29 21 DAML file

O nt oPl an O nt oPl an

Semantic Web

DAML file DAML file DAML file Planning Agent Domain Viewer Plan Viewer Goals & constraints Plans

Ontology files

2003-8-29 22

O nt oPl an O nt oPl an:

: Domain & Plan Viewers Domain & Plan Viewers

slide-12
SLIDE 12

2003-8-29 23

Application Application

  • A consultation system for computer configurations

A consultation system for computer configurations

1. 1.

To suggest compatible To suggest compatible hardware components hardware components

2. 2.

To meet user preference To meet user preference

  • Hardware Ontology:

Hardware Ontology: hardware hardware-onto.

  • nto.daml

daml

  • Distributed product information (from manufacturers)

Distributed product information (from manufacturers) are are located in distributed instance files, as well as in located in distributed instance files, as well as in local sources local sources

H ar dwar e H ar d di s k Di s pl ay Car d Soundc ar d

<<s ubcl as s >> <<s ubcl as s >>

M

  • t he r boar d

CPU RAM

<<s ubcl as s >>

M

  • ni t or

<<s ubcl as s >>

2003-8-29 24

Example of Operators Example of Operators

Ac t i on: BuyCPU( ?x) Pr e c ondi t i ons : CPU( ?x) , H ave M

  • t he r boar d( ?y) ,

pl at f or m ( ?x, ?s oc ke t ) , c puPl at f or m ( ?y, ?s oc ke t ) , s ys t e m Bus Spe e d( ?x, ?f s b) , m axFSB( ?y, ?f s b) Ef f e c t s : addi t i ons : H ave CPU( ?x) de l e t i ons :

Socket of the motherboard fits the CPU

Socket of the motherboard fits the CPU

FSB of the motherboard is compatible with the bus

FSB of the motherboard is compatible with the bus speed of the CPU speed of the CPU

slide-13
SLIDE 13

2003-8-29 25

WI Challenge #2 WI Challenge #2

Distributed Agents + Coordination Distributed Agents + Coordination

slide-14
SLIDE 14

2003-8-29 27

Example: Distributed Scheduling Example: Distributed Scheduling

Given: Given:

A group of people, each

A group of people, each

  • f whom has specific
  • f whom has specific

available available time slots time slots

A set of

A set of constraints constraints among people (e.g., among people (e.g., persons A and B will not persons A and B will not be present at the same be present at the same time) time)

Find: Find:

An available time slot

An available time slot when when all constraints are all constraints are satisfied satisfied

2003-8-29 28

Satisfiability Satisfiability Problems (SAT) Problems (SAT)

SAT SAT Meeting scheduling Meeting scheduling Clause Clause Constraint Constraint Variable and its domain Variable and its domain Person and time slots Person and time slots Literals in a clause Literals in a clause Persons involved in a Persons involved in a constraint constraint CNF: CNF:

? = = ‘and and’, , V = = ‘or

  • r’,

, Ci Ci = clause, clause, Lji Lji = literal literal (variable or its negation)

(variable or its negation)

slide-15
SLIDE 15

2003-8-29 29

Local Search Local Search

(Selman, Levesque, and Mitchell, 1992;

Selman, Levesque, and Mitchell, 1992; Gu Gu, 1992) , 1992)

Take the solution space

Take the solution space (i.e., the Cartesian (i.e., the Cartesian product of all variable product of all variable domains) as a search domains) as a search space, and search it space, and search it based on a certain rule based on a certain rule

Randomly select one

Randomly select one position as the start point position as the start point to search to search

At each step, move to a

At each step, move to a neighboring position neighboring position according to according to the rule (i.e., the rule (i.e., heuristic) heuristic) UnitWalk UnitWalk (2002) (2002) SDF (2001) SDF (2001) R-Novelty+ Novelty+ (1999) (1999) Novelty+ Novelty+ (1999) (1999) R-Novelty Novelty (1997) (1997) Novelty Novelty (1997) (1997) GWSAT/ GWSAT/ Tabu Tabu (1997) (1997) WalkSAT WalkSAT/Ta Ta bu bu (1997) (1997) GWSAT GWSAT (1994) (1994) WalkSAT WalkSAT (1994) (1994) GSAT (1992) GSAT (1992)

2003-8-29 30

If

If

people are

people are distributed distributed in in different places different places

the time slot

the time slot information is information is NOT NOT centralized centralized

Then

Then

centralized local

centralized local search methods search methods become become ineffective ineffective How to solve it?

slide-16
SLIDE 16

2003-8-29 31

Multi Multi-

  • Agent SAT (

Agent SAT (M

ASSAT M ASSAT)

)

(with X. J in) (with X. J in)

Use multiple computational agents

Use multiple computational agents

Decompose the search space into several

Decompose the search space into several sub sub-spaces spaces

Each agent decides how to locally search a

Each agent decides how to locally search a sub sub-space (i.e., its environment) space (i.e., its environment)

Through

Through local local interactions interactions between agents and between agents and their environments, the agents coordinately find their environments, the agents coordinately find a a global global solution solution to the given problem to the given problem

2003-8-29 32

Formulation Formulation

Divide variables into

Divide variables into u groups groups

Agent

Agent a i : variable group : variable group Vi = { = { v i 1

i 1, …

, v i k

i k}

Environment

Environment ei of

  • f ai: Cartesian Product of the variable group,

: Cartesian Product of the variable group, Di = = D

i 1 i 1× …

×Di k

i k

Agent

Agent’ s position s position e i j : : j th

th value combination in

value combination in Di = = Di 1

i 1× …

× Di k

i k

The position combination of all agents

The position combination of all agents { e 1k… , , e i

j , …

} : } : a possible solution a possible solution

Basic move strategies of agent

Basic move strategies of agent a i : : ψi : D : D

1× …

× D

i × …

× D

u —

> Di

Best

Best-move: move: ψi - bes t

bes t =

= e i

j ’ ,

, s t s t , , e v al e v al ( e i j ’ ) = M ax ) = M ax e v al e v al ( e i k) ) ( f or al l ( f or al l e i k∈ D

i )

Better

Better-move: move: ψi - bet t er

bet t er =

= e i

j ’ ,

, st st , , eval eval ( e i j ’ ) > ) > e v al e v al ( e i j )

Random

Random-move: move: ψi - r andom

r andom =

= e i j ’ , , e i j ’ i s a r andom pos i t i on i n i s a r andom pos i t i on i n Di

slide-17
SLIDE 17

2003-8-29 33

MASSAT Procedure MASSAT Procedure

For For i =1 i =1 to to M AX M AX- Cy c l e s Cy c l e s If all clauses are satisfied Then return If all clauses are satisfied Then return T; For For all distributed agents all distributed agents Select one (or more) of three behaviors Select one (or more) of three behaviors {Best {Best-move, Better move, Better-move, or Random move, or Random- move}; move}; Perform selected behavior(s); Perform selected behavior(s); EndFor EndFor If no agent moves Then If no agent moves Then Modify the weights of unsatisfied Modify the weights of unsatisfied clauses clauses ; End End Update Update T according to new positions of according to new positions of agents; agents; EndFor EndFor

2003-8-29 34

Comparison Based on Time Step Comparison Based on Time Step

– Benchmark SAT problem packages from SATLIB – Time step is the minimum unit

  • At each time step of SDF,
  • nly one variable is

flipped

  • In M

ASSAT, agents move asynchronously

1 time step MASSAT

Var v T F F T

Local Search Based Methods

slide-18
SLIDE 18

2003-8-29 35

MASSAT: MASSAT: Variable Variable-

  • Based Agent Representation

Based Agent Representation

Nodes: Each agent represents a variable in an SAT problem Edges: If two variables appear in a common clause, there will be an edge between the agents corresponding to these two variables, indicating these two agents need to coordinately assign their respective values Nodes: Each agent represents a variable in an SAT problem Edges: If two variables appear in a common clause, there will be an edge between the agents corresponding to these two variables, indicating these two agents need to coordinately assign their respective values

2003-8-29 36

Clause Clause-

  • Based Agent Representation

Based Agent Representation

Nodes: Each agent represents a clause in an SAT problem Edges: If clauses represented by two agents have a common literal, there will be an edge between these agents, which indicates these two agents need to coordinate Nodes: Each agent represents a clause in an SAT problem Edges: If clauses represented by two agents have a common literal, there will be an edge between these agents, which indicates these two agents need to coordinate

slide-19
SLIDE 19

2003-8-29 37

Agent Networks Agent Networks

  • How does the

How does the topology topology of an agent

  • f an agent

network reflect the network reflect the complexity complexity of solving

  • f solving

distributed SAT? distributed SAT?

  • Do agent networks formed in

Do agent networks formed in M ASSAT M ASSAT have have small small-world world topologies? topologies?

Small Small-World World

1. 1. Highly clustered Highly clustered 2. 2. The shortest path The shortest path length between length between any two nodes is any two nodes is small small

  • Jin and Liu (AAMAS

Jin and Liu (AAMAS’03) 03) have experimentally have experimentally proven that proven that constraint constraint satisfaction in a satisfaction in a small small- world world is inefficient is inefficient

WI Challenge #3 WI Challenge #3

Social Networks + Self Social Networks + Self-

  • Organization

Organization

slide-20
SLIDE 20

2003-8-29 39

Social Networks Social Networks

A

A Social Network Social Network comprises a group of comprises a group of people with a pattern of interactions people with a pattern of interactions among them among them

A Social Network

A Social Network is a is a self self-organizing

  • rganizing

structure structure of users,

  • f users,

information, and information, and communities of communities of expertise expertise or

  • r

practice practice

2003-8-29 40

Self Self-

  • Organization: Game of Life

Organization: Game of Life

(Conway, 1970)

`Life

Life’ rules rules are applied to an initial are applied to an initial population of live cells (i.e., black circles) population of live cells (i.e., black circles)

slide-21
SLIDE 21

2003-8-29 41

Social Networks + Self Social Networks + Self-

  • Organization:

Organization:

Game of Game of RoboNBA

RoboNBA (with C. H. Ng)

(with C. H. Ng)

Live cells

Live cells

  • `Life

`Life’ rule rule

  • Environment

Environment

Patterns

Patterns

  • Distributed

Distributed player agents player agents

Decision/strategy

Decision/strategy

Virtual court

Virtual court

NBA

NBA-like like games games

Basketball competition is a complex behavior: Basketball competition is a complex behavior:

  • Team players interact

Team players interact locally

locally

  • It is difficult to predict a

It is difficult to predict a complete

complete match

match

2003-8-29 42

Example: Example: Moving Moving Action (1 of 8) Action (1 of 8)

blue circle

blue circle – defender defender’s s

  • ld position
  • ld position

red circle

red circle – offender

  • ffender’s

s

  • ld position
  • ld position

blue square

blue square – defender defender’s s new position new position

red square

red square – offender

  • ffender’s

s new position new position

slide-22
SLIDE 22

2003-8-29 43 2003-8-29 44

www. www.nba nba.com .com

slide-23
SLIDE 23

2003-8-29 45

RoboNBA RoboNBA Games

Games

The teams used for testing the accuracy of

The teams used for testing the accuracy of the system: the system:

Team Name Team Name Rank in NBA Rank in NBA Wins Wins

Dallas Maverick

Dallas Maverick 1st 1st 0.75% 0.75%

Philadelphia

Philadelphia Sixers Sixers 10th 10th 0.592% 0.592%

Washington Wizards

Washington Wizards 20th 20th 0.461% 0.461%

Cleverland Cavs

Cleverland Cavs 29th 29th 0.197% 0.197%

2003-8-29 46

Dallas Mavericks VS Philadelphia Dallas Mavericks VS Philadelphia Sixers Sixers (5 matches) (5 matches)

Mavericks win Mavericks win Sixers Sixers 4 4 – 1 (averages 90.6 1 (averages 90.6 – 79.2) 79.2)

slide-24
SLIDE 24

2003-8-29 47

Dallas Mavericks VS Philadelphia Dallas Mavericks VS Philadelphia Sixers Sixers (real) (real)

Total TO: 10 Total TO: 10 Team Team Rebs Rebs: 3 : 3 100.0% 100.0% 54.5% 54.5% 49.4% 49.4% 107 107 7 10 10 7 22 22 24 24 39 39 30 30 9 11 11-11 11 12 12-22 22 42 42-85 85 240 240 TOTAL TOTAL DNP DNP Popeye Jones Popeye Jones 0-0 0-0 0-1 2 Avery Johnson Avery Johnson 1 2 1 1 0-0 0-0 0-1 7 Antoine Antoine Rigaudeau Rigaudeau 4 0-0 0-0 0-2 7 Raef LaFrentz Raef LaFrentz 1 2 2 0-0 0-0 0-2 12 12 Adrian Griffin Adrian Griffin 11 11 2 2 1 1 0-0 3-3 4-6 13 13 Walt Williams Walt Williams 16 16 1 1 3 3 2 2 2-2 4-5 5-8 22 22 Nick Van Nick Van Exel Exel 8 5 2 2 1 9 5 4 0-0 0-0 4-7 24 24 C SHAWN BRADLEY SHAWN BRADLEY 30 30 1 3 1 2 14 14 12 12 2 4-4 2-5 12 12-19 19 35 35 F DIRK NOWITZKI DIRK NOWITZKI 2 1 2 5 2 2 0-0 0-0 1-5 41 41 F RAJA BELL RAJA BELL 21 21 1 1 3 13 13 4 3 1 3-3 2-5 8-20 20 37 37 G STEVE NASH STEVE NASH 19 19 2 1 1 5 3 2 1 2-2 1-4 8-14 14 40 40 G MICHAEL FINLEY MICHAEL FINLEY

PTS PTS BS BS TO TO ST ST PF PF AST AST TOT TOT DEF DEF OFF OFF FTM FTM

  • A

3GM 3GM-A FGM FGM-A MIN MIN PO PO S

PLAYER PLAYER REBOUNDS REBOUNDS

MAVERICKS MAVERICKS

Mavericks win Mavericks win Sixers Sixers 107 107-94 94

Source: www.nba.com

2003-8-29 48

RoboNBA RoboNBA Games

Games

14 14 – 16 16

(69.3 (69.3 – 74.3) 74.3)

(real: 93 (real: 93-84) 84)

Wizards Wizards

(left) (left)

24 24 – 6

(75.4 (75.4 – 69.0) 69.0)

(real: 116 (real: 116-103) 103)

18 18 – 12 12

(78.7 (78.7 – 76.5) 76.5)

(real: 88 (real: 88-80) 80)

Sixers Sixers

(left) (left)

25 25 – 5

(82.7 (82.7 – 72.9) 72.9)

(real: 114 (real: 114-93) 93)

24 24 – 6

(86.0 (86.0 – 73.5) 73.5)

(real: 106 (real: 106-101) 101)

22 22 – 8

(87.0 (87.0 – 83.7) 83.7)

(real: 107 (real: 107-94) 94)

Mavericks Mavericks

(left) (left)

Cavs Cavs

(right) (right)

Wizards Wizards

(right) (right)

Sixers Sixers

(right) (right)

TEAMS TEAMS

(30 matches) (30 matches)

slide-25
SLIDE 25

Issues and Directions Issues and Directions

2003-8-29 50

  • 1. Goal-directed

Services

  • 6. Meta-knowledge

(planning control)

  • 7. Semantics

(relationships)

  • 2. Personalization

(identity)

  • 5. Coordination

(global behavior)

  • 4. PSML

(representation)

  • 8. Association

(roles)

  • 3. Social &

psychological context (sensitivity)

  • 9. Reproduction

(population)

  • 10. Self-aggregation

(feedback)

Top 10 Top 10

Issues Issues

(best means/ends)

slide-26
SLIDE 26

2003-8-29 51

The Wisdom Web The Wisdom Web

To enable humans to gain

To enable humans to gain practical practical wisdom wisdom of

  • f living

living, , working working, and , and playing playing Wisdom

Wisdom: ( : (Webster Dictionary Webster Dictionary, page 1658) , page 1658) The quality of being wise; knowledge, and the The quality of being wise; knowledge, and the capacity to make due use of it; capacity to make due use of it; knowledge of the knowledge of the best ends and the best means best ends and the best means

To provide

To provide a

a medium medium for for knowledge and experience (e.g., knowledge and experience (e.g.,

  • f the G
  • f the Grand Canyon

rand Canyon) sharing ) sharing

a

a supply supply of self

  • f self-organized resources for driving
  • rganized resources for driving

sustainable sustainable knowledge creation knowledge creation and and scientific or social scientific or socialdevelopment/evolution development/evolution

2003-8-29 52

Summary Summary

Background

Background

Challenges

Challenges

Semantic Web + Planning:

Semantic Web + Planning: Ont oPl an

Ont oPl an

Distributed Agents + Coordination:

Distributed Agents + Coordination: M

ASSAT M ASSAT

Social Networks + Self

Social Networks + Self-Organization: Organization: RoboNBA

RoboNBA

Issues and Directions

Issues and Directions

The Wisdom Web

The Wisdom Web

slide-27
SLIDE 27

2003-8-29 53

The End The End