Dynamic Agent Communities Facilitating to Distant Learning in a - - PDF document

SLIDE 1

1

Dynamic Agent Communities Facilitating to Distant Learning in a Virtual University Information Space

Vadim Ermolayev

• Dept. of Mathematical Modelling and Information Technologies,

Zaporozhye State University, 66, Zhukovskogo st., 330600, Zaporozhye, Ukraine, tel/fax:+380 61 264 17 24, E-mail: eva@zsu.zaporizhzhe.ua

EVA: IS2000-VUDE, 06.11.2000 2

Motivation and Background

Layered mediator University IS with unified interfaces Hierarchy of distributed wrapped legacy IS and information resources I guess I know how to use this!!! It’s like doing my everyday work, but electronically!!!

VUIS VUIS is considered to be a passive media inhabited by active “beings” (agents) representing real life actors…

It’s like CA Real World

• r Microsoft Encarta

SLIDE 2

2

EVA: IS2000-VUDE, 06.11.2000 3

Active VUIS Inhabitants

Persistent MAS and Member Agents

S U R [ \ S U R [ \

Generic MAS agents

Middle Agent Middle Agent Ontology Agent SDS Coordination Agent Cloning Agent

University MAS

Proxy Agent(s)

VU MAS V Dept. MAS Lower level V Dept. MAS Proxies “wrap” respective MAS and are the representative members in the higher level MAS

EVA: IS2000-VUDE, 06.11.2000 4

Agents’ Interactions – Parametric Feedbacks

)) ( ~ ),..., ( ~ ( ~ ~

1

X y X y Y Y

n

=

Ai Aj

) , ( Y X f a =

1.0 0.2 0.4 0.6 0.8 5 10 15 No of students at the class Professor’s interest in new students

− ) ( ~

1 X

y

Entry test difficulty level

− ) ( ~ X yn Parametric feedbacks – expressed attitude, capability to perform requested action (policy) at a certain state

a – accept a new student

to the class

SLIDE 3

3

EVA: IS2000-VUDE, 06.11.2000 5

Agent Communities

as dynamic coalitions for performing tasks (sets of works) Why? – there are MAS already

• tasks emerge and are soon accomplished
• agents join the coalition if they are capable

to contribute to the task execution

• an agent may participate in more than one coalitions

at a time How? – accept or reject the work, react with parametric feedbacks What is changing in time?

• the coalition – agents join and move out
• agent’s state – capability to perform a work
• agent’s knowledge (beliefs) about the capabilities
• f the other MAS members

EVA: IS2000-VUDE, 06.11.2000 6

Framework Models

Generic Agent Model Communication Model Coordination Model Functional System / Component Model Process Model

???? Evolution Model ????

SLIDE 4

4

EVA: IS2000-VUDE, 06.11.2000 7

Agent’s proactive self-development and self-adaptation in response to the changes in MAS and environment caused by task(s) execution

Agent’s Evolution Model

Capabilities Beliefs

B A1 An Ai

             

j n w w A j

c X c c

j j i

... ) ( ...

1

j

w

[ ]

1 ,

Si Sj

) , ( ~ X s f Y

i

←

Y ~

)} ( ), ( ), ( { F t F q X r s

A A i =

) (

A

X r

• parameter constraints

) (

A

F q

• policy constraints

) (F t

• transition function

A j i

S s s ∈ ,

EVA: IS2000-VUDE, 06.11.2000 8

Case Study – VUDE Domain

Professor (PRA), Assistant (AA), Course master (CMA), Librarian (LA) agents External influences and MAS reactions

Department MAS

Middle Agent Task communities Influences Facilitator cloning Ontology queries Co-ordination Middle Agent Middle Agent Tutor Agent (User) Tutor Agent (User) Ontology Agent

Proxy Agent

SDS Coordination Agent Cloning Agent Proxy Agent (secretary)

PhD recruiting scenario: Phase 1: A PhD candidate submits the CV

and indicates his/her intention to become a PhD student (PA, CA, TA)

Phase 2: The CV is analysed and the best

Professor Match is searched (TA, PRA-s)

Phase 3: Qualified candidate passes the test

from the chosen professor (TA, PRA, PA)

Phase 4: Successful candidate is interviewed

and assigned to a research project (TA, PRA)

Phase 5: The professor and his assistant

prepare the individual curriculum for the accepted candidate as well as the list of recommended reading (TA, PRA, AA, CMA, LA)

SLIDE 5

5

EVA: IS2000-VUDE, 06.11.2000 9

VDept Agents and their Roles

(w.r.t. the Case Scenario)

PA

L

Z aporozhye State U niversity

• Inst. of M athem atics and Cybern etics

E lena Vetukh Vyachyslav Tolok Sergey B orue Vadim E rmo layev Sergey G om enyuk Vitaly M ukhin Nadezhda M a tv eyshina Nataly a K eberle Vladim ir D avidovsky M axim Shum akov Victor G ristcha k Anatoly D oroshenko S ecretary M rs C hair Prof. PR 1 Prof. PR 2 Prof. PR 3 A ss. A 1 A ss. Prof A 2 A ss. Prof A 3 A ss. Prof A 4 A ss. Prof A 5 A ss. Prof A 6 Visiting: Prof. PR 4 Virtual: Prof. P R 5 Faculty: P rof

• Dept. O f M athem atical M odelling and IT

Proxy P

Virtual F aculty

A 2 A 3 A 4 A 5 A 6 P R 2 P R 3 P R 4

PRA5 PRA1 AA1 CM A PRA1 LA

TA

Utility Agents: CA – Clone a TA each time a new PhD candidate comes to PA COA – Monitor agents activities, Accept work results to its SDS, Provide work params on request. OA – Provide common dept.

• ntologies on request

Middle Agents: PRA1-PRA5 – evaluate candidates’ qualifications; provide a test, evaluate test results, arrange the interview, provide course recommendations AA1-AA6 – prepare working plan, prepare curriculum CMA – provide electronic courses, issue calls for new courses LA – provide electronic textbooks, issue calls for new teaching materials Proxy Agent: PA – accept outer influences,

• rder to clone TA, pipeline the candidate to his TA

EVA: IS2000-VUDE, 06.11.2000 10

Phase 3. Testing, Step 1 – Test is ordered.

Modelling ( How did they play their roles? ):

TA :

n

t PRAi

CoA

COA

No Results

t

TA

1

: execute w

TA

W } , { ~

g TA d TA TA

W W W =

∅

} , , (

1 1 1

Y X t Requre_Tes w = ), , , ' _ ('

2 2 2

Y X Test Provide w = ), , , ' ('

3 3 3

Y X Test w = ), , , ' Re _ ('

4 4 4

Y X sults Evaluate w =

)} , , ' _ ('

5 5 5

Y X Marks Analyse w =

∅

No works for redirection TA: Require_Test policy: 1. Generate 4 works 2. Invoke TA to influence PRAi with TA itself with

5 2

w w −

4 2,w

w

5 3,w

w Just monitoring…

SLIDE 6

6

EVA: IS2000-VUDE, 06.11.2000 11

), , , ' Re _ ('

4 4 4

Y X sults Evaluate w = TA

5 3

w w

PRAi

TA PRAi

CoA

COA

Results:

t

Phase 3. Testing, Step 2 – Test is provided.

Modelling ( How did they play their roles? ):

:

n

t

1

: execute w

TA

W :

1 + n

t

4 2

w w

g TA

W

delay

5 3,w

w

execute, delay

2

w

4

w

} _ { ~2 > =< = FILENAME Form Test Y

i

PRA

Monitoring and accepting feedback(s)…

), , , ' ('

3 3 3

Y X Test w =

)} , , ' _ ('

5 5 5

Y X Marks Analyse w = redirect to itself

TA

redirect to itself

PRAi: Provide_Test policy: 1. Generate test file ( ) according to the ontology and received params X2 2. Invoke PRAi to influence COA with: (‘Accept_Results’, )

2

~

i

PRA

Y

2

~

i

PRA

Y

EVA: IS2000-VUDE, 06.11.2000 12

Phase 3. Testing, Step 3 – Test is performed.

Modelling ( How did they play their roles? ):

TA

5 3

w w

PRAi

TA PRAi

CoA

COA

Results:

t

:

n

t

1

: execute w

TA

W :

1 + n

t

4

w

delay

5 3,w

w

execute, delay

2

w

4

w

} _ { ~2

3

> =< = = FILENAME Form Test Y X

i

PRA

Monitoring, accepting feedback(s), providing work parameters… ) , , ' Re _ ('

4 4 4

Y X sults Evaluate w =

)} , , ' _ ('

5 5 5

Y X Marks Analyse w =

redirect to itself

TA

redirect to itself

TA :

2 + n

t

execute, delay

3

w

5

w

PRAi Params:

TA: Perform_Test policy: 1. Require and accept parameters from COA 2. Invoke human user to fill in the test form. 3. Invoke TA to influence COA with: (‘Accept_Results’, )

3

~

TA

Y

3

w

} _ { ~3 > =< = FILENAME Results Test YTA delay

4

w

SLIDE 7

7

EVA: IS2000-VUDE, 06.11.2000 13

Phase 3. Testing, Step 4 – Test results are evaluated.

Modelling ( How did they play their roles? ):

TA

5

w

PRAi

TA

CoA

COA

Results:

t

:

n

t

1

: execute w

TA

W :

1 + n

t

4

w

delay

5 3,w

w

execute,delay

2

w

4

w

} _ { ~3

4

> =< = = FILENAME Results Test Y X

TA

Monitoring, accepting feedback(s), providing work parameters… )} , , ' _ ('

5 5 5

Y X Marks Analyse w = TA

redirect to itself

TA :

2 + n

t

execute,delay

3

w

5

w

PRAi Params: )} ,... ( ), ,... {( ~

1 1 4 k k PRA

s s m m Y

i =

delay

4

w :

3 + n

t TA

delay

5

w

PRAi execute

4

w

EVA: IS2000-VUDE, 06.11.2000 14

Phase 3. Testing, Step 5 – Parametric Marks are analysed.

Modelling ( How did they play their roles? ):

TA

PRAi

TA

CoA

COA

t

:

n

t

1

: execute w

TA

W :

1 + n

t

delay

5 3,w

w

execute,delay

2

w

4

w

Monitoring, providing work parameters… )} , , ' _ ('

7 7 7

Y X failure

• n

Inform w = TA

Negative: inform

TA :

2 + n

t

execute,delay

3

w

5

w

PRAi )} ,... ( ), ,... {( ~

1 1 4 5 k k PRA

s s m m Y X

i =

= Params: delay

4

w :

3 + n

t TA

delay

5

w

PRAi execute

4

w

5

w

TA :

4 + n

t

execute

5

w

g TA

W TA

Positive: Interview – Phase 4.

)} }, , { , ' _ ('

6 5 1 6 6

Y X X X Interview Require w = =

No Results

∅

SLIDE 8

8

EVA: IS2000-VUDE, 06.11.2000 15

Summary

What has been done: What has been not, yet:

The processes of DE may be modelled and executed as tasks without pre-defined task plans in frame of the formal agent-based approach to modelling and management of the processes

• f information interchange.

The particularities of the framework:

• generic agent skeletons,
• parametric feedbacks,
• agents ability to evolve

provide for the scalability and pro-active dynamic character

• f the approach.

Planned for short-term:

• Architectural framework on the base
• f the presented formal approach
• Semantic issues: Ontology representation,

interoperability, knowledge sharing among generic agents, within MAS and task communities.

Planned for middle-term:

• Design and implementation of the prototype

for a Virtual University Department

• Further evaluation of the applicability
• f the approach to VUDE, BPM, DLIB

domains