Download Mutlidimensional Detective Alfred Inselberg Streeable

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
Mutlidimensional Detective
Alfred Inselberg
Streeable, Progressive, Mutlidimensional
Scaling
Matt Williams, Tamara Munzner
Rylan Cottrell
Mutlidimensional Detective
• Transformation of multivariate relations
into 2-D patterns
• A discovery process for visual data
mining
Parallel Coordinates
•
•
Visualize without loss of information.
Properties
•
•
•
•
•
Low complexity. # Dimensions = # Variables
Works for any # of dimensions
Variables treated uniformly
N - dimensional Object
•
•
Recognized under projective transformations.
Conveys information on the properties
Based on rigorous math/algo results.
DON’T PANIC
Data
• 473 batches of processors
• 16 variables
• X1 - % of yield
• X2 - quality
• X3 ... X12 - are different types of
defects
• X13 ... X16 - denote a physical
parameter
Maximize yield and quality
Batches with the highest
quality
Portion of Slovenia
Satellite Data
B1..B5, B7 - Intensity of reflected electromagnetic wavelengths
B6 - Intensity of emitted thermal IR from object
X,Y - Map Position
Portion of Slovenia
Multidimensional
Scaling
• Create a low dimensional layout of data
• Distance between points best
represents the points in higher
dimensional data.
Steerable,
Progressive MDS
• Problem - No Interactive exploration of
high-dimensional data sets
• Unreasonable time cost associated
data sets that are large in dimensions
and points
• Steering - focuses computational power
Layout a random subset of the data set
Divide bin in two
Apply high-dimensional distance
A new random subset of points are added into the layou
Focus is placed on user defined bin
A new subset of random points selected from the
unplaced points in the selected region are added
The process is repeated as the user
refines his selection
Standard Layout
(Morrison)
MDSteer
50,000 data points
http://www.cs.ubc.ca/~tmm/papers/mdsteer/videos/MDSteer1.mov
Standard Layout
(Morrison)
MDSteer
40,000 data points
http://www.cs.ubc.ca/~tmm/papers/mdsteer/videos/MDSteer2Combined.mov
References
•
•
•
•
Alfred Inselberg: The Automated Multidimensional Detective. INFOVIS 1997: 107-114
Alfred Inselberg: Parallel Coordinates: Visuak Multidimensional Geometry and its Applications. 2004.
http://www.math.tau.ac.il/~aiisreal/index_files/lect-pdf/lect-intro.pdf
Matt Williams, Tamara Munzner: Steerable, Progressive Multidimensional Scaling. INFOVIS 2004:
57-64. Project website http://www.cs.ubc.ca/labs/imager/tr/2004/mdsteer/
Matt Williams: QuestVis and MDSteer: The Visualization of High-Dimensional Environmental
Sustainability Data. MSc. Thesis. 2004.
Related documents