cse 564: visualization introduction - stony brook...
TRANSCRIPT
CSE 564: Visualization
Introduction
Kl M llKlaus Mueller
Computer Science Department
Stony Brook University
The Visual Brain
O 50% f th h b i i d di t d t i i dOver 50% of the human brain is dedicated to vision and visual representations,• decoding visual information• high-level processing of visual information• thinking with visual metaphors
Input Device: The Eye Sensor: The Cones and Rods
T t f t tiTwo types of receptors on retina: rods and cones
Rods:Rods:• spread all over the retinal surface (75 -
150 million)• low resolution no color vision but very• low resolution, no color vision, but very
sensitive to low light (scotopic or dim-light vision)
Cones:Cones:• a dense array around the central
portion of the retina, the fovea centralis (6 7 million)(6 - 7 million)
• high-resolution, color vision, but require brighter light (photopic or bright-light vision)bright light vision)
Wiring: The Visual Pathways Processing Unit: The Visual Cortex (V1, V2)
Vi l t b k i t i t diff t tVisual cortex breaks input up into different aspects: • color, shape, motion, depth
LGN= Lateral Geniculate Nucleus
Pre-Attentive Processing
If t it t f t l d t t dIf you want it or not: some features are always detected
And fast – within 200 ms or less
Pre-Attentive Processing
Wh i it f t?Why is it so fast?
Well, because 50% of the brain is dedicated to vision
Vision is a MASSIVELY parallel processor dedicated to• detect
l• analyze• recognize• reason with
visual input
Pre-Attentive Processing
S iti it t diff iSensitivity to differences in:• color, orientation, size, shape, motion, shading, 3D depth, …
Pre-Attentive Processing
B t th li it j ti d ’t k llBut there are limits: conjunctions don’t work well
quick: find the blue circle
Pre-Attentive Processing
S f t / t th thSome features/cues are stronger than others:
Pre-Attentive Processing
W d tt hi h f t tt ti f tWords are patterns, which form strong pre-attentive features• this would have been different if this had been done in Arabic
There are limits, however• let’s see the next experiment p
Pre-Attentive Processing
R di 1Reading 1
Aoccdrnig to a rscheearch at an Elingsh
uinervtisy, it deosn't mttaer in waht oredr the
ltteers in a wrod are, the olny iprmoetnt tihng
is taht frist and lsat ltteer is at the rghit pclae.
The rset can be a toatl mses and you cany
sitll raed it wouthit porbelm. Tihs is bcuseae
we do not raed ervey lteter by it slef but thewe do not raed ervey lteter by it slef but the
wrod as a wlohe
Pre-Attentive Processing
Now, is tihs ture? Raed on….
Pre-Attentive Processing
R di 2Reading 2
Anidroccg to crad cniyrrag lcitsiugnis
planoissefors at an uemannd, utisreviny
in Bsitirh Cibmuloa, and crartnoy to the
duoibus cmials of the ueticnd rcraeseh,
a slpmie, macinahcel ioisrevnn ofp ,
ianretnl cretcarahs araepps sneiciffut to
csufnoe the eadyrevy oekoolnrcsufnoe the eadyrevy oekoolnr
Pre-Attentive Processing
R di 2Reading 2
According to card carrying linguistics
professionals at an unnamed, university
in British Columbia, and contrary to the
dubious claims of the uncited research,
a slpmie, macinahcel ioisrevnn ofp ,
ianretnl cretcarahs araepps sneiciffut to
csufnoe the eadyrevy oekoolnrcsufnoe the eadyrevy oekoolnr
Pre-Attentive Processing
R di 2Reading 2
According to card carrying linguistics
professionals at an unnamed, university
in British Columbia, and contrary to the
dubious claims of the uncited research,
a simple, mechanical inversion ofp ,
internal characters appears sufficient to
confuse the everyday onlookerconfuse the everyday onlooker
What To Learn From This
Th h i l t (HSV) t l t ( i l) iThe human visual system (HSV) tolerates (visual) noise very well• it can read the randomly garbled text very well• machines (equipped with computer vision) are poor at this
Humans have only limited computational capacityh d t t fi d l t d i h t t• hard to execute a fixed rule to decipher text
• especially once the text gets longer (7±2 rule of working memory)• this is where computers excel
The fact that computers deal poorly with noisy patterns is exploited in CAPTCHA• CAPTCHA: Completely Automated Public Turing Test to tell• CAPTCHA: Completely Automated Public Turing Test to tell
Computers and Humans Apart• used to ensure that an actual human is interacting with a system• some examples:some examples:
- creating a new gmail or yahoo account (prevent spammer accounts)- submitting files, data, email
CAPTCHA
CAPTCHA i d tl di t t d tt th t diffi ltCAPTCHA: noisy and vastly distorted patterns that are difficult to recognize by machines
Completely Automated Public Turing testto tell Computers and Humans Apart
CAPTCHA: One More
A t i i l ith h bAs computer vision algorithms have become more sophisticated at CAPTCHA character recognition…• the latest approach is object recognition
Visualization Examples Visualization Examples
Visual Analytics
A i fi ldA new emerging field
Visual Analytics = the science of reasoning with visual informationinformation
Pairs machine intelligence (computing, bit-representations) with human intelligence (creativity, visual representations)g ( y, p )
Recent Headline
Connecting The Dots… Connecting The Dots…
Connecting The Dots…
structured data unstructured data
other data…
Connecting The Dots…
structured data unstructured data
other data… hypothesis:- connect by number
Connecting The Dots…
structured data unstructured data
other data… hypothesis:- connect by number
It’s Not Always That Easy….
More data- structured
- unstructured
Will the hypothesis workWill the hypothesis work here?
It’s Not Always That Easy….
More data- structured
- unstructured
Will the hypothesis workWill the hypothesis work here? No…
It’s Not Always That Easy….
More data- structured
- unstructured
It’s Not Always That Easy….
More data- structured
- unstructured
Slowly sculptSlowly sculpt..
It’s Not Always That Easy….
More data- structured
- unstructured
Slowly sculptSlowly sculpt..
other data…
He is white
It’s Not Always That Easy….
More data- structured
- unstructured
Slowly sculptSlowly sculpt..
It’s Not Always That Easy….
More data- structured
- unstructured
Slowly sculptSlowly sculpt..
Typical Data
salient feature
salient data point
outlier
Connecting The Dots: Cause + Effect
Observe data infer causality relations
Connecting The Dots: Cause + Effect
Simple causal chain …. complex interactions
Connecting The Dots: Cause + Effect
Complex causal network … yet vastly simplified (mind map)
Connecting The Dots: Cause + Effect
Very complex causal network
Data
Oft t i b dOften present in abundance• what is relevant/irrelevant?• under what conditions/task?
Data
Oft t i b dOften present in abundance• what is relevant/irrelevant?• under what conditions/task?
May be missing in critical areas• where are these areas?
d k h t d ’t k ?• do we know what we don’t know?
Data
Oft t i b dOften present in abundance• what is relevant/irrelevant?• under what conditions/task?
May be missing in critical areas• where are these areas?
d k h t d ’t k ?• do we know what we don’t know?
Prone to uncertainties• acquisition errors noise ambiguities misinterpretationsacquisition errors, noise, ambiguities, misinterpretations, …
Data
Oft t i b dOften present in abundance• what is relevant/irrelevant?• under what conditions/task?
May be missing in critical areas• where are these areas?
d k h t d ’t k ?• do we know what we don’t know?
Prone to uncertainties• acquisition errors noise ambiguities misinterpretationsacquisition errors, noise, ambiguities, misinterpretations, …
Complex interactions• not just “A causes B”: A Bj• but multivariate: wA1⋅A1 + wA2⋅A2 + … B
Data Interaction Model
Data Interaction Model
Connects the dots• add structure to the data
Data Interaction Model
Connects the dots• add structure to the data
information
Traditional Data Analytics: Human
dl i td t deepanalysis
pre-analysis reportdata
Traditional Data Analytics: Machine
dl i td t deepanalysis
pre-analysis reportdata
Traditional Data Analytics: Either/Or
dl i td t deepanalysis
pre-analysis reportdata
Comparison: The Good
Human MachineMemory Tremendous
Computational Bandwidth Highp gEndurance HighIntelligence High
Creative Thinking HighCreative Thinking HighPattern Recognition High
Comparison: The Bad
Human MachineMemory Limited
Computational Bandwidth PoorpEndurance LowIntelligence Limited
Creative Thinking LimitedCreative Thinking LimitedPattern Recognition Poor
Comparison: Complementing Strengths
Human MachineMemory Limited Tremendous
Computational Bandwidth Poor Highp gEndurance Low HighIntelligence High Limited
Creative Thinking High LimitedCreative Thinking High LimitedPattern Recognition High Poor
Visual Analytics: Exploiting Synergies
dl i td t deepanalysis
pre-analysis reportdata
Visual Analytics: Exploiting Synergies
sense-making loopsense making loop
dl i td t deepanalysis
pre-analysis reportdata
Human-Machine Analytical Discourse
E bl d h / hi i tiEnable good human/machine communication• key to synergistic problem solving
Human-Machine Analytical Discourse
E bl d h / hi i tiEnable good human/machine communication• key to synergistic problem solving
Visual communication has best bandwidth• highly parallel and noise-tolerant
Human-Machine Analytical Discourse
E bl d h / hi i tiEnable good human/machine communication• key to synergistic problem solving
Visual communication has best bandwidth• highly parallel and noise-tolerant
Visual story telling with data & findings• needed: visual language rooted in data analyticsneeded: visual language rooted in data analytics
Human-Machine Analytical Discourse
E bl d h / hi i tiEnable good human/machine communication• key to synergistic problem solving
Visual communication has best bandwidth• highly parallel and noise-tolerant
Visual story telling with data & findings• needed: visual language rooted in data analyticsneeded: visual language rooted in data analytics
Visual interface 2-way communication channel• interaction is key• gesture-driven human/machine discourse
Visualization of 1-3 Dimensional Data
2D i l f( )
1D signal f(x)
2D signal f(x, y)
4D i l f( ti )
2D signal, shown as height field
3D signal f(x, y, z)4D signal f(x, y, z, t=time)example: 3D heart in motion
High-Dimensional Data
C id th li t f t fConsider the salient features of a car:• miles per gallon (MPG)• top speed• acceleration• number of cylinders• horsepower• weight• year• country originy g• brand• number of seats• number of doors• reliability (average number of breakdowns)• and so on...
H d i ll l ti f ?How do we visually compare a selection of cars?• this is a high-dimensional visualization problem… how to do this?
High-D Visualization Example: Parallel Coordinates
a car as a 7-dimensional data point
High-D Visualization Example: Parallel Coordinates
a database of cars
High-D Visualization Example: Parallel Coordinates
after some clustering
High-D Visualization Example: Parallel Coordinates
with mean trend
Optical Illusions
C t th bl k d t !Count the black dots!
Optical Illusions
Optical Illusions Optical Illusions
Optical Illusions Optical Illusions
Are the purple lines straight or bent?
Optical Illusions
Which circle in the middle is bigger?
Optical Illusions
Do you see gray areas in between the squares? Now where did they come from?
Optical Illusions
You should see a man's face and also a word...
Hint: Try tilting your head to the right, the world begins g , g
with 'L'
Optical Illusions: Sidewalk Art
Julian Beever
Optical Illusions: Sidewalk Art
Julian Beever
Optical Illusions: Sidewalk Art
Julian Beever
Optical Illusions: Sidewalk Art
Julian Beever
Optical Illusions: Sidewalk Art
Julian Beever
Optical Illusions: Sidewalk Art
Julian Beever
Explanation
( ti )
real-world view
screen (retina)
focal point
drawn illusiondrawn illusion (equally perceived)
The Power of the Visional System
S th h i l t (HSV) i t f t b t it’So the human visual system (HSV) is not perfect, but it’s extremely powerful
Vision is an integral part of life
Vision is the gateway to higher-level regions of the brain
Exploit this fast and powerful processor for• complex data analyses, creative tasks, communicating ideas
The science of visualization