Partiview

1
Partiview
Visually exploring high-dimensional data
Dinoj Surendran Computer Science Department
University of Chicago
Stuart Levy Experimental Technologies Group
National Center for Supercomputing Applications
University of Illinois at Urbana-Champaign

• Open source downloadable freeware
• Industrial strength smoothly interact with 4d
  data on a laptop
• Millions of points
• - Thousands of images (graphics memory size)
• Can use pictures (jpegs) as glyphs
• 3d coordinates are weighted combinations of N
  data attributes
• can change weights in real-time
• Stereo red-cyan, GeoWall, etc
• Windows / Linux / OS X
• Traditionally used for presenting astronomy data
  to public (Hayden/AMNH, Adler Planetarium, Sloan
  Digital Sky Survey, Cosmus Group _at_ UChicago)
• Useful for machine learning researchers making
  dim-reduction clustering algorithms because
• can color points by attributes (can change which
  attribute in real-time)
• pictures at points feature is useful for data
  with natural visual representation good
  qualitative error analysis
• can partition data into groups
• usable Matlab interface
• can make cute demos!

Handwriting Recognition
MNIST dataset of handwritten digits after
dimension reduction by Laplacian Eigenmaps
(Belkin Niyogi 03)
Face Recognition
ORL database of 400 faces, top three PCA
coordinates
Image Similarity
Galaxy Classification
Corel Image Database, after LPP (He Niyogi 03)
Galaxies from the Sloan Digital Sky Survey after
applying Laplacian Eigenmaps to information about
their spectral type.