[PPT] - Delft Cooperation on Intelligent Systems The Human in the loop: a PowerPoint Presentation

SLIDE 1

Delft Cooperation on Intelligent Systems

The Human in the loop: a crisis-continuation factor?

Christiaan Huygens Symposium 13 March 2008

Delft Cooperation on Intelligent Systems

SLIDE 2

2

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Introduction

The speaker The D-CIS Lab

SLIDE 3

3

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

3

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

D-CIS Lab strategy

Integrated Framework Consortium integration

JPA

Joined Program Activities Partners’ portfolio Start of the project

2003 2004/5 2006

Expanding the partnership: UT, UM, RU, RUG, Shell,..

Integrated Framework Consortium integration

JPA

Joined Program Activities Partners’ portfolio Start of the project Integrated Framework Consortium integration

JPA

Joined Program Activities Partners’ portfolio Start of the project

2003 2004/5 2006

Expanding the partnership: UT, UM, RU, RUG, Shell,..

2007/2008 2005/2006 2003/2004

SLIDE 4

4

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

4

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Preferred Application Domain

Crisis Management

Knowledge & Resource management Collaborative Decision Making in Chaotic Environments Human Cognitive Performance Factors Sustained Team Effectiveness Controlling Emergence

Real-time Chaotic Complex S-o-S Information intense Smart OODA Humans in the loop

SLIDE 5

5

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

5

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Applied Research: Demand Articulation & Solution Provisioning = Transillitation Business Units AUAV/swarming A A valorisatiegebied

SLIDE 6

6

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Overview The context: crisis management New paradigm for information systems: AAC Mixing human information and expertise, sensor data and structured knowledge to improve understanding and decision making The influence of human cognitive performance factors Some questions for the audience

SLIDE 7

7

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

The context: A crisis taxonomy

Different types of strategies and demands

Man-made disaster Natural disaster Pandemic disaster

Different types of strategies and demands

Man-made disaster Natural disaster Pandemic disaster

SLIDE 8

8

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

The context: crisis management

Crisis: an unwanted situation, caused by some form of disturbance, that affects one of several of our vital infrastructures Basic crisis management strategy: find and eliminate the cause, restore normal processes and rebuild ‘old’ situation.

SLIDE 9

9

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

9

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Field of Applied Research

Actor-Agent Communities (AACs) are complex self-

rganizing systems that are used for creative and

proactive problem solving in dynamic and unpredictable environments. The main application of AACs is to find optimized solutions for otherwise intractable problems, for example in areas such as command & control and collaborative decision making. They contain two distinct types of interacting information processing components: people (the actors) and software agents (the agents) that proactively collaborate, guided by a shared interest. These AAC systems comply with the theoretical concepts and technical specifications described in the Actor-Agent Community Paradigm and supporting documents, published by the D-CIS Lab, the research partnership of Thales Nederland, Delft University of Technology, TNO and the University of Amsterdam

ACTOR ACTOR-

AGENT COMMUNITIES

AGENT COMMUNITIES       

SLIDE 10

10

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Why mixing information?

!"

#"

SLIDE 11

11

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Charateristics of a crisis in development

$% %!

&%' % (% !"! % % %!!)"!

SLIDE 12

12

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Challenges

%

&%! *"+' “” !!,%!

!%“” -%!

& !!."!

!

/0%%%% ! 1% !

SLIDE 13

13

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Modeling with Bayesian Networks

23%" ! ""!!!%%% ε 3% ,#"! $"! 2%%! &! 4!% %! & ! 35" !

) , ( ˆ max arg ) | ( ˆ max arg ε ε

i h i h class

h P h P h

i i

= =

SLIDE 14

14

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Causal Probabilistic Models: Example

GasX CondC SC1

1

Smell Sympt SC1

2

SC2

3

EC1

1

EC1

m

EC1

m+1

EC1

n

T H EC2

1

EC2

E4

1

E4

p

E5

1

E5

q

C1 C2 C3

3

C2

3

C1

3

Each node is directly connected with a small fraction of other variables Factorization of the JPD over the variables consists of factors which can be computed independently.

SLIDE 15

15

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Distribution of models and fusion processes

SLIDE 16

16

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Distributed Perception Networks (DPN)

. 35%&&'! ! .0%35 ,! 1%" ,! („#%06$ 70!“ #%

#85%

% !"!

SLIDE 17

17

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Something about robustness

335 % %!

!

('!

!"# $ % $ &' 9!: % 9!; &( !"# $ % $ &) 9!7 % 9!< &* + ,+

SLIDE 18

18

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

True vs. Modelled Probability Distributions

4%% %%

350! / ! &81&81" (#%&&'""

/35 % !!!

SLIDE 19

19

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

True Probability Distributions vs. CPT Parameters

1% &81 (% &81% %9!=! 1% &81-./ !1 %%!4 %4&

0.6 0.2 e2 0.4 0.8 e1 c2 c1 P(E|C) 0.99 0.45 e2 0.01 0.55 e1 c2 c1 P(E|C)

> > < <

0.8 0.1 e2 0.2 0.9 e1 c2 c1 P(E|C)

> <

+ ,+

SLIDE 20

20

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

Questions to be answered

/ 6

/!(

!
%&&)"!
!

SLIDE 21

Human Cognitive Performance Factors

21

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems

++0+ +1

80%4-$"

SLIDE 22

26

3-

T h e i n f l u e n c e

f

m

d
n

c h a n g e b l i n d n e s s

22

SLIDE 23

26

3-

T h e i n f l u e n c e

f

m

d
n

c h a n g e b l i n d n e s s

23

& % 8% 2

3

6

3

SLIDE 24

26

3-

T h e i n f l u e n c e

f

m

d
n

c h a n g e b l i n d n e s s

24

& /

32

SLIDE 25

26

3-

T h e i n f l u e n c e

f

m

d
n

c h a n g e b l i n d n e s s

25

6

%

3

SLIDE 26

26

3-

T h e i n f l u e n c e

f

m

d
n

c h a n g e b l i n d n e s s

26

8 2 2 ! 2 2

8>%

1+

SLIDE 27

26

3-

T h e i n f l u e n c e

f

m

d
n

c h a n g e b l i n d n e s s

27

8 2+ %

4

SLIDE 28

26

3-

T h e i n f l u e n c e

f

m

d
n

c h a n g e b l i n d n e s s

28

1%

799!" &

8%
"+1

SLIDE 29

26

3-

T h e i n f l u e n c e

f

m

d
n

c h a n g e b l i n d n e s s

29

?@A 8@A

SLIDE 30

26

3-

T h e i n f l u e n c e

f

m

d
n

c h a n g e b l i n d n e s s

30

35

2

%

B

BCD!9=

SLIDE 31

th e

/

%%
!

SLIDE 32

th e

%
%

%% $% %

SLIDE 33

th e

1

"!1

% ,

SLIDE 34

th e

SLIDE 35

th e

5 & # !!" #

1+

SLIDE 36

th e

8%!

%%!"
%
!!

%" % +% !

31+

SLIDE 37

th e

-!

(

4%
%

,.

3

SLIDE 38

th e

4%

%
4
1

%

3.

+

SLIDE 39

End

39

2 6 / 3 / 2 8

Delft Cooperation on Intelligent Systems