An Information-theoretic Framework for Visualization Min Chen and Heike Jänicke Swansea University and Universität Heidelberg

1 Motivation: A Theoretic Framework „ Quantitative measurement and reasoning „ Explanation of facts and phenomena „ Laws and guidelines for optimization „ Falsifiable predictions Looking for theories? Lookingfortheories?

1 Motivation 2 Communication n & Visualization 3 Quantifying Visual Info plot 4 Explanation n - logarithmic & redundancy 5 Laws – interaction & user study

6 Prediction n – DP inequality & ... 7 Conclusions & Future Work

2 Communication and Visualization „ Is a visualization

system a communication system?

Example Example: avisualcommunicationsystem imagefrom: http://chicagodesignintern.blogspot.com/

A General Visualization Pipeline (without interaction) raw data D

Source

N

information I

N

geometry & labels G

Visual Mapping

Filtering

image V

N

Rendering

optical signal S

N

Optical Transmission

Displaying

image V'

N Viewing

N

optical signal S'

N

information I'

Perception

N

knowledge K

Cognition

Noise message M

Source

signal S

Encoder (Transmitter)

signal S'

Channel

message M'

Decoder (Receiver)

A General Communication System

Destination

Destination

Three Visualization Subsystems raw data D

Source

information I

N Filtering

N

geometry & labels G

Visual Mapping

image V

N

Rendering

vis-encoder image V

N

optical signal S

N

optical signal S'

Optical O ti l Transmission

Displaying

image V'

N

Viewing

vis-channel image V'

N

information I'

Perception

N

knowledge K

Cognition

Destination

vis-decoder message M

Source

signal S

Encoder (Transmitter)

N Channel

signal S'

message M'

Decoder (Receiver)

A General Communication System

Destination

Three Visualization Subsystems raw data D

image V

N visencoder

Source

message M

Source

Encoder (Transmitter)

N

image V'

vischannel

visdecoder

?

?

signal S

signal S'

N Channel

compactness

N

knowledge K

Destination

message M'

Decoder (Receiver)

error detection error correction

A General Communication System

Destination

3 Quantifying Visual Information „ Random variable

X „ It takes values

x1 , x2 ,  , xm „ Probability mass function

p ( xi ) „ Entropy

H(X )

m

¦ p ( xi ) log 2 p ( xi ) i

ClaudeE.Shannon (19162001)

minimal 64 pixels 256

Entropy: Example „

minimal 2 256 pixels

192

128

„

Ti Time Series S i z

64 independent samples

z

each sample is an integer in [0, 255]

z

probability mass function is i.i.d. (independent and identically identically-distributed) distributed)

Entropy

H (Z ) 64

„

0

8 16 24 32 40 48 56 64

1 1 ¦ ¦ log 2 256 t 0 i 0 256

512

Time Series Plot z

0

64 255

Minimal 256x64 p pixels ((214 p pixels))

minimal 64 pixels

76543210

256

byte 1

X X X

X X X X X X X X X X X X X X X X X X

The Most Compact, Compact Why Not?

X X

X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

byte 16

minimal 2 256 pixels

192

128

byte y 32

byte 48

64

byte 64

0 0

8 16 24 32 40 48 56 64

X X X X X

minimal 8 pixels

„

Th mostt compact: The t z

„ minimal 6 64 pixels

X X X X X X

One pixel per bit seems not enough for vision z

„

64 bytes (512 bits)

4x4 pixels per bit o 213 bits

Not a counter example: z

Sequential or parallel

z

Salient information o at o

Three Visualization Subsystems raw data D

image V

N visencoder

Source

perceptual efficiency cognitive load

vischannel

?

signal S

signal S'

Encoder (Transmitter)

N Channel

compactness

N

knowledge K

visdecoder

intuitiveness clarity

message M

Source

N

image V'

Destination

message M'

Decoder (Receiver)

error detection error correction

A General Communication System

Destination

Three Measures for Visualization „

Entropy of Input Data Space: H(X)

„

Visualization Capacity: V(G)

„

Di l C Display Capacity: it D cf. (Yang-Peláez & Flowers, 2000)

Visual Mapping Ratio (VMR)

V (G ) H(X )

Information Loss Ratio (ILR)

max( H ( X )  V (G ),0) H(X )

V ((G G) Display Space Utilization (DSU) D

4 Explanation - Logarithmic Plot & Redundancy „ Logarithmic plot z

Why is it a useful in some situations?

z

In what situation?

z

What does it try to optimize?

„ Redundancy y in Visualization Design g z

Good and bad?

z

When is it useful?

z

Should this community pay more attention to “redundancy”?

minimal 64 pixels 256

Information Loss Ratio (ILR) „

192

Di l S Display Space R Restriction t i ti

minimal 2 256 pixels

z

128

information loss:

25% 256

64

8 16 24 32 40 48 56 64

Evenly distributed probability mass function

„

Linear visual mapping

„

ILR is a p probabilistic measure about z

a data space X

128

z

not an instance xi

0 0

„

192

64

0

64x64 pixels

0

8 16 24 32 40 48 56 64

(a) evenly distributed p

Non uniform Distribution Non-uniform „

Li Linear visual i l mapping i probability

D: information loss:

information loss:

25.0%

25.8% C:

256

256

192

192

B: 128

p

p

p

128

64

64

A A: 0

0 0

8 16 24 32 40 48 56 64

(a) evenly distributed p

0

8 16 24 32 40 48 56 64

(b) unevenly distributed p

p

1 8 1 8 1 4 1 2

Z

linear

Z'

256

64

224

56

192

48

160

40

128

32

96

24

64

16

32

8

0

0

Non uniform Distribution Non-uniform N li Nonlinear visual i l mapping i

„

information loss:

information loss:

information loss:

25.0%

25.8%

22.6%

256

256

256 192

192

192

128

128

128

64

64

64

0

0 0

8 16 24 32 40 48 56 64

(a) evenly distributed p

0

8 16 24 32 40 48 56 64

(b) unevenly distributed p

0 0

8 16 24 32 40 48 56 64

(c) 4 regional mappings

D:

C:

B:

A:

p

1 8

p

1 8

p p

1 4

k of

1 2

p

1

p

1

p

1

2k

Non uniform Distribution Non-uniform

2 k 1 ......

p

1

p

1

p

2k

„

23

L Logarithmic ith i visual i l mapping i

22 1 2

information loss:

information loss:

information loss:

information loss:

25.0%

25.8%

22.6%

0%

256

256

28

256 192

192

192

128

26

128

128

64

24

64

64

0

0 0

8 16 24 32 40 48 56 64

(a) evenly distributed p

22

0

8 16 24 32 40 48 56 64

(b) unevenly distributed p

20

0 0

8 16 24 32 40 48 56 64

(c) 4 regional mappings

0

8 16 24 32 40 48 56 64

(d) logarithmic plot

Three Visualization Subsystems raw data D

image V

N visencoder

Source

vischannel

?

signal S

signal S'

Encoder (Transmitter)

N Channel

compactness

N

knowledge K

visdecoder

intuitiveness clarity low info. loss

message M

Source

N

image V'

Destination

message M'

Decoder (Receiver)

error detection error correction

A General Communication System

Destination

Three Visualization Subsystems raw data D

image V

N visencoder

Source

vischannel

N

error detection error correction

signal S

signal S'

Encoder (Transmitter)

N Channel

compactness

knowledge K

visdecoder

intuitiveness clarity low info. loss

message M

Source

N

image V'

Destination

message M'

Decoder (Receiver)

error detection error correction

A General Communication System

Destination

300

£M 271

262

How?

250 199

200

141

SMrt grp p

GreenG grp p

FoodXP grp p

2008

138

SMrt grp p

50

GreenG grp p

100

FoodXP grp p

150

200

„

Redundancy y z

see Rheingans & Landreth, 1995

„

Display Space Utilization (DSU) is generally very low

„

25-50 fps refresh rate can be utilized

2009

5 Laws - Interaction and User Studies Law 3: The information about an overview in one of its detailed view is the same as that about that detailed view in the overview.

Example 1 Mutual Information

p(x,y)

Hint

No Hint

Vortex

25%

25% negative

No V t Vortex

5%

45%

false positive

false

I ( X ,Y ) „

p ( x, y ) ¦ ¦ p( x, y ) log 2 p( x) p( y ) xX yY

Example 1: I = 0.147

Example 1 Mutual Information

p(x,y)

Hint

No Hint

Vortex

25%

25% negative

No V t Vortex

5%

45%

false

false positive

Example 2 p(x,y)

Hint

No Hint

Vortex

40%

10% negative

No V t Vortex

10%

40%

false positive

false

„

E Example l 1 1: I(X;Y) I(X Y) = 0.147 0 147

„

Example 2: I(X;Y) = 0.278

The Role of User Studies „

Do quantitative D tit ti measurements make user studies less important?

Example 1 p(x,y)

Hint

No Hint

Vortex

25%

25%

No Vortex

5%

45%

Example 2 p(x,y)

Hint

No Hint

Vortex

40%

10%

No Vortex

10%

40%

If we can measure p(x,y), p(x y) p(x), p(x) p(y), p(y) ... „

E Example l 1 1: I(X;Y) I(X Y) = 0.147 0 147

„

Example 2: I(X;Y) = 0.278

6 Prediction – DP Inequality & a Prediction „ Data processing inequality z “No clever manipulation of data can improve the inferences that can be made from the data” data [Cover and Thomas, 2006] „ Markov chain conditions z Closed coupling: (X, Y), (Y,Z) z X and Z are conditionally independent d d

p(x, y, z) = p(x) p(y|x) p(z|y) p(x)

p(y|x)

X

Process 1

p(z|y) Y

I (X; Y)

Process 2

Z

I (Y; Z)

I ( X ;Y ) t I ( X ; Z ) interaction U1 X

Process 1

interaction U2 Y

Process 2

Z

„ What if the condition is broken? domain knowledge about X

I ( X ;Y ) t I ( X ; Z )

X

Process 1

Y

Process 2

Z

IEEE CG&A CG&A, Jan Jan. 2009 „

I t Interactive ti visualization i li ti

„

Information-assisted visualization

„

Knowledge-assisted visualization

7 Conclusions and Future Work „ The process of gaining insight can be

i improved db by using i numbers. b „ Information theory can explain many

phenomena in visualization visualization, z

but be careful with naive use.

„ The current limitations: z

assumption of memoryless, z lack of real p(xi). „ This talk is an overview of the paper. „ Details of related work, z

in particular, Matt Ward’s suggestion (see Purchase et al. 2008)

RichardW.Hamming (19151998)

“The purpose of computing is insight, not numbers.”

The Role of a Theoretic Framework Facts W W W W W W W W W W W W W W W W W W W W

T

T

T

T

T

T

T

T

T

T

W T

W T

Wisdom

W T

T

8T T T

Theory Theoretic Framework

Th nkk You Thank u !

An Information-theoretic Framework for Visualization Min Chen and Heike Jänicke Swansea University and Universität Heidelberg

An Information-theoretic Framework for Visualization

Is a visualization system a communication system? Example: a visual communication system image from: http://chicagodesignintern.blogspot.com/ ...

2MB Sizes 2 Downloads 194 Views

Recommend Documents

NetTopo: A Framework of Simulation and Visualization for ... - CiteSeerX
Oct 30, 2008 - hazardous situations [3], deploying a testbed is unwanted since ... in a physical WSN testbed and interact with each other to ..... The private variable is initialized by calling the ..... an easy way for users to configure their input

Two-stage Framework for Visualization of Clustered ...
Georgia Institute of Technology. 266 Ferst Drive, Atlanta, GA .... scatter matrix Sw, the between-cluster scatter matrix Sb, and the total (or mixture) scatter matrix St ...

NetTopo: A Framework of Simulation and Visualization for ... - CiteSeerX
Oct 30, 2008 - networks (WSNs) applications drive the fast development of research in various .... e.g., energy consumption, bandwidth management. NetTopo.

An Architectural Framework for Interactive Music Systems
Software Architecture, Interactive Systems, Music soft- ... synthesis of data media of different nature. ... forms (e.g. Max/MSP [19] and Pure Data [24]), and oth-.

AN EVIDENCE FRAMEWORK FOR BAYESIAN ...
generalization, and achieve desirable recognition performance for unknown test speech. Under this framework, we develop an EM iterative procedure to ...

NetTopo: A framework of simulation and visualization ...
MAC layer protocol, they can simulate extremely large networks. Packet level. ..... the storage file with format formed by serializing the Vir-. tualWSN can be ...

An Explanation of Cryptographic Key Fingerprint Visualization ...
The authenticity of host '[target.net]:23. ([69.141.191.69]:226)' can't be established. ECDSA key fingerprint is. 1f:0d:00:0c:bd:ba:cb:a0:39:75:ee:91:7d:16:d1:fe.

An Extended Framework of STRONG for Simulation ...
Feb 29, 2012 - Indeed, STRONG is an automated framework with provable .... Construct a local model rk(x) around the center point xk. Step 2. .... We call the sample size required for each iteration a sample size schedule, which refers to a.

Instrumentino: An open-source modular Python framework for ...
Official Full-Text Paper (PDF): Instrumentino: An open-source modular ... 1. Introduction. In the process of scientific research, many laboratories around ..... [18] N. Barroca, et al., Wireless sensor networks for temperature and .... The communicat

An Event-based Framework for Characterizing the ...
for evolving networks, based on our framework. Categories and Subject Descriptors: H.2.8 Database. Management: Database Applications - Data Mining.

Zemberek, an open source NLP framework for Turkic ...
source, platform independent NLP framework not only for Turkish but also all Turkic languages. In. This paper .... There is no agreed standard naming scheme defined for. Turkish yet. Most suffixes contain a production ..... environment: AMD Athlon 64

An Empirical Framework for Automatically Selecting the Best Bayesian ...
Keywords: Bayesian networks; Data mining; Classifi- cation; Search ... In deciding which classifier will work best for a given dataset there .... The software used to ...

IBPM: An Open-Source-Based Framework for InfiniBand ... - GitHub
Evaluation of Computer and Communication Systems (MMB) and. Dependability and ... 2 These authors are with science+computing ag, Tuebingen, Germany.

Reference Framework for Handling Concept Drift: An ...
In predictive analytics, machine learning and data mining the phenomenon ...... [13] A. Bifet, R. Gavalda, Learning from time-changing data with adaptive.

An island-model framework for evolving neuro ...
showing that the aforementioned framework can parallelise the controller design .... will outline ongoing work into the application of migration to more complex ..... http://www.cnes.fr/web/5719-msl-09-at-a-glance.php. [28] D. P. Miller and T. L. ...

An Argumentation-based Framework for Deliberation in ...
eration policy, and the counterargument generation policy are case-based tech- niques. For join deliberation .... like generation and selection of arguments and counterarguments. In our approach, the ...... and languages that support argumentation, b

An Edge-preserving Filtering Framework for Visibility ...
scene depth information [3] are rather limited because of the acquisition of additional information. Recently, ... fulfil the demands of real-time applications. Based.

An Integrated Framework for Checking Concurrency ...
a programs runtime behavior based on analyzing its source code [8]. The strength of static analysis is that it can con- sider all possible behaviors of a program.

An Integrated State- and Event-Based Framework for ...
Control software development plays an increasingly greater role in the .... in a distributed way, per functional aspect, such as, paper path, printing process, etc.

An Empirical Framework for Automatically Selecting the Best Bayesian ...
In deciding which classifier will work best for a given ... ing rationally the one which performs best. ..... The fact that this degree of accuracy was achieved on a.