Role-Based Collaboration and E-CARGO · 1 Role-Based Collaboration and E-CARGO Haibin Zhu, PhD (朱海滨)Professor and Coordinator Computer Science Program Director, Collaborative

1

Role-Based Collaboration and E-CARGO

Haibin Zhu, PhD (朱海滨)Professor and CoordinatorComputer Science ProgramDirector, Collaborative Systems LaboratoryNipissing University, 100 College Dr., North Bay, ON P1B 8L7, [email protected]://www.nipissingu.ca/faculty/haibinzSponsors: IBM, NSERC, OPIC, and DRDC-Canada

2

Contents Collaboration Role-Based Collaboration The E-CARGO Model

Role Negotiation Agent Evaluation

Role Assignment Role Transfer Role Playing

How to Discover How to Verify Conclusion

3

Collaboration

Collaboration Collaboration is an activity/process that

involves more than one agent to participate in. To be skilful in collaboration requires a life-long

time to learn and practice. A team is a group of people who are mutually

dependent on one another to achieve a common goal.

Without collaboration, there would be no team.

4

Collaboration made easy/efficient!

-The goal of E-CARGO -Through modeling and algorithms

5

Challenge!

2013 Survey

6

https://www.pinterest.ca/pin/331296116308388145/

Why do we do research on collaboration?

GoalCollaboration made easy and efficient!

PropertiesCollaboration is required!

Collaboration is challenging!Collaboration is valuable!

Collaboration is interesting!

7

Nature of Collaboration

(Task) Distribution Task specification, Agent evaluation,

Task assignment, Optimization. (Task) Execution [(Co-)Operation] Process, Sharing, Negotiation,

Interaction, Coordination, Decision making.

8

Why collaboration is hard?

9

Distribution Co-Operation(Execution)

Distribution and execution are interleaved!

If we divide them clearly:

10

Task ExecutionTask Distribution

Task Distribution Task Execution

11

Role-Based Collaboration (RBC)(Standing on the Shoulders of Giants)

Giants: Object-Orientation, Agent Systems, Distributed systems, Chinese Culture and Philosophy

12

Why RBC?“名不正，则言不顺；言不顺，则事不成。”

“Ming bu zheng, ze yan bu shun; yan bu shun, ze shi bu cheng”.

“If terminology is not corrected, then what is said cannot be followed. If what is

said cannot be followed, then work cannot be accomplished. ”

----Confucius, Lun Yu: Zi Lu, No. 13, Section 3.

Role Specification （角色定义/规范）。

Role Theory (Confucius) 君君臣臣父父子子。

“The king is king, and the minister is minister, the father is father, and the son is son.”

Meaning: Good government obtains only when all the relative duties are maintained.

Distribution.

13

What is RBC?

Role-based collaboration (RBC) is

a computational methodology that mainly uses

roles as underlying mechanisms to facilitate collaboration activities

(abstraction, classification, separation of concerns, dynamics,

coordination and interactions).

142006

15

RBC and E-CARGO have been investigated for >15 years!

What is RBC?

Role-based collaboration (RBC) is emerging into

an investigative methodology that mainly uses

roles as the key componentto model and formalize problems

in collaboration and complex systems.

162016

RBC is such a research topic to deal with complex systems

Distribution, The king is the king: Role Assignment

[Co-]Operation: Role Execution Right titles: Role Specification Collectivism, Decision Making, Optimization and

tradeoff , Save against a rainy day: Group Role Assignment (GRA), GRA+

Dynamics, Survival of the fittest: Adaptive Collaboration, Roles as Team Dynamics

Multiplayer game: Role Assignment, Execution

17+Combinatorics

(Co-)OperationExecution

Distribution

Flowchartof

RBCThe user of this chart is the manager!

The Model

19

The E-CARGO model ∑ ::= <C, O, A, M, R, E, G, s0, H> Environments Classes Agents Roles Groups Objects

20

Object-Oriented (C, O)天时 The Universe

Agent-Oriented (C,O,A,G)天时+人和The Universe + The People

Role-Based(C,O,A,R,E,G)天时地利人和The Universe + The Earth+The People

21

The views on an agent

An agent is playing roles.A role is the interface for an agent to interact with

others.

22

E-CARGO explains Shakespeare

All the world’s a stage (E, C, O),And all the men and women are merely

players(A);They all have their exits and

entrances(G);And one man in his time plays many

parts(R). -----W. Shakespeare , As You Like It, Act II, Scene 7

23

24

The Initial Internal Elements of E-CARGO (2006)

H. Zhu, and M. Zhou, “Role-Based Collaboration and its Kernel Mechanisms,” IEEE Trans. on Systems, Man, and Cybernetics, Part C: Applications and Reviews, vol. 36, no. 4, 2006, pp. 578-589.

The Revised Internal Elements of E-CARGO (2015)

25

H. Zhu, “Adaptive Collaboration Systems”, IEEE Systems, Man, and Cybernetics Magazine, vol. 1, no. 4, Oct. 2015, pp. 8-15.

C, O No change!

26

Role Assignment

Major Symbols/Concepts

m – the number of agents, (i =0, …, m-1) --A n – the number of roles, (j =0, …, n-1) --R L – the vector of lower ranges of roles(vs. U)-- E Q – the matrix of agent evaluation --G T – the assignment matrix – working G

By combing with C and O , we can investigate more!

27

28

The Q Matrix: result of role negotiation and agent evaluation

r0 r1 r2 r3

a0a1a2a3a4a5

Workable roles and group T is a matrix, T[i, j] belongs to{1, 0} T[i, j]=1 means agent i is assigned to role j,

otherwise not. L[j] expresses how many agents are required

by role j Role j is workable if ∑ , Group g is workable when all its roles are

workable.

29

30

L = [2 1 1 2]

A qualification matrix Q and an assignment matrix T

Algorithm Exhaustive Search

Approximately O(m!), not practical After adapting the Kuhn-Munkres

Algorithm O(m3)

31

1

0 ][

n

j jLm

GRACAR/G Group Role Assignment with Conflicting

Agents on Roles / in a Group. New matrix Ac

Ac[i1, i2]=1 means that agents i1 and i2are in conflict.

That agents i1 and i2 are in conflict means that agents i1 and i2 cannot be assigned with the same role (GRACAR) or in the same group (GRACAG).

32

33

1

0

1

0],[],[max

m

i

n

jjiTjiQ

}1,0{],[ jiT )0,0( njmi

][],[1

0jLjiT

m

i

)0( nj

1],[1

0

n

ijiT )0( mi

)0,,,0(1]),[],[(],[ 21212121 njiimiijiTjiTiiAc

Group Role Assignment with Conflicting Agents on Roles (GRACAR/G) is to find a workable T to

subject to

)2,10,,,0(1]),[],[(],[ 2121221121 njjiimiijiTjiTiiAc

)1(

)2(

)3(

/)4(

)5(

How to solve

34

IBM ILOG CPLEX Optimization Package (CPLEX), Matlab, Maple, ……

Our contribution? Problem formalization; Estimation of the problem complexity; A practical solution based on CPLEX; Conditions for feasible solutions; Guidelines for using such a solution; and Confirm common sense by simulations.

35

La[i] expresses how many roles can be assigned to by agent i

RBC is an innovative discovery methodology in engineering research.

GMRA-Group Multi-Role Assignment

36

1

0

1

0],[],[

m

i

n

jjiTjiQ

1} {0,],[ jiT )0,0( njmi

][],[1

0

jLjiTm

i

)0( nj

][],[1

0iLjiT a

n

j

)0( mi

GMRA is to find a workable T to max σ=subject to

(1)

(2)

where (1) is a 0-1 constraint; (2) makes the group workable; (3) makes an agent be assigned with a limited number of roles.

(3)

Solutions: CPLEX or KMB

Role Transfer (RT) is required when initial assignment is

done and dynamic evaluation is not available.

It can be solved by group role assignment, called role re-assignment, where only some of agents needs to change their roles.

37

38

Role transfer problems

Current roles

Potential roles

Not workable group

Workable group

Complexity Exhaustive Search

Approximately O(m!), not practical Adapted Kuhn-Munkres Algorithm O(m3)

Temporal Role Transfer(Strong Restriction) is still complex even with the Kuhn-Munkres Algorithm.

39

1

0 ][

n

j jLm

Y. Sheng, H. Zhu, X. Zhou, and W. Hu, “Effective Approaches to Adaptive Collaboration via Dynamic Role Assignment”, IEEE Trans. on Systems, Man, and Cybernetics: Systems, vol. 46, no. 1, Jan. 2016, pp. 76 - 92.

40

GRA+Time->DGRA

How to Discover by RBC and E-CARGO?

41

General Coverage (RBC) + Special Points (E-CARGO) Role Negotiation in A, R, E

Agent Evaluation + special requirement in A, R, E

Role assignment/transfer + special conditions in A, R, E GRACAR/G, GMRA

Role Execution: Simplify the problems of Multi-Agent Systems and solve them.

42

GRA=>GRA+

(GRACAR/G, GMRA, GRACCF, GRAB, GRABC)(Ac, La , Ccf , Pa , B)

43

How to verify?-Simulation-Comparison

44

How to compare (Method 1)?(Team Performance)

45

])[(],[],['1

0

1

03 realGRAjiTjiQ

m

i

n

j

])[(],[],[1

0

1

01 idealGRAjiTjiQ

m

i

n

j

)(],['],[1

0

1

02

GRAjiTjiQm

i

n

j

332 /)(

Example: (GRACAR)

46

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 117.6

17.8

18

18.2

18.4

18.6

18.8

19

Effect

Gro

up P

erfo

rman

ce

GRAGRACAGRACR

m =30, n =10, pc =10%, 1≤L[j] ≤3.

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 114

14.5

15

15.5

16

16.5

17

Effect

Gro

up P

erfo

rman

ce

GRAGRACAGRACR

m =30, n =6, pc=10%, 1≤L[j]≤5.

1

23

1

23

47

H. Zhu, “Avoiding Conflicts by Group Role Assignment”, IEEE Trans. on Systems, Man, and Cybernetics: Systems, vol. 46, no. 4, April 2016, pp. 535-547.

How to compare (Method 2)?(Solution efficiency)

1. Find necessary/sufficient conditions for a GRA+ problem;

2. Design algorithms to check the necessary/sufficient conditions;

3. Set up simulation cases; 4. Compare with CPLEX by solving

100 problems.

48

Example: Theorems for GMRA Theorem 1: The necessary condition for the

GMRA problem to have a feasible solution is that

Theorem 2:

49

1

0

1

0][][

n

j

m

i

a jLiL

.

GMRA (with Q, T, L, and La) has a feasible solution if and only if there exists an integer q, such that Lq is empty and for each i, Li

* | Li-1Lai |

(0 < i q).

Comparison with ILOG CPLEX solutions

50

The times used by processing 100 problems (n=m/2,1≤L[j]≤20, and 1≤La[j]≤10)

H. Zhu, D. Liu, S. Zhang, S. Teng, and Y. Zhu, “Solving the Group Multi-Role Assignment Problem by Improving the ILOG Approach”, IEEE Trans. on Systems, Man, and Cybernetics: Systems, vol. 47, no. 12, Dec. 2017, pp. 3418-3424.

51

RBC & E-CARGO could contribute to or be contributed: Optimization

More problems are discovered with E-CARGO Use Opt methods to solve RBC problems

Multi-Agent Systems Combine Collectivism/Individualism

Parallel Computing Parallel algorithms for RBC problems. Solving RBC problems may provide solutions to the key problems of

parallel computing, e.g., load balance, resource management, and CPU scheduling.

Cutting-Edge Technologies IoT, Smart City, Social Networking, Social Simulation, Scheduling,

Adaptive Systems, AI () …

Health Human Body Simulation/Teamwork for diagnosis and treatment

52

Conclusion

结语

53

54

Collaboration =

Task Distribution +

(Co-)Operation (Task Execution)

The greatest is the simplest! Keep it simple!

RBC is… A computational methodology A discovery methodology

55

E-CARGO is the model for RBC

Negotiation, Evaluation, Assignment, and Execution

m, n, L, Q, T, Ac, La, Ccf, P, B, i, j

56

Question?

[email protected]