Semisupervised Classification of Astronomical Objects

1
Semisupervised Classification of Astronomical
Objects

• FY05 Highlights
• Completed Project, June 2006
• Published Astrophysical Journal article detailing
  methods and results
• Presented poster at 2006 AAS meeting in
  Washington, DC
• Follow-on Plans
• Pursuing applications in other areas planetary
  science, Earth science
• Extending method to higher dimensional data sets

• David Bazell (PI)
• Johns Hopkins University Applied Physics
  Laboratory
• David.Bazell_at_jhuapl.edu
• David Miller, Penn State University
• Project Summary
• We investigated methods to classify
  astronomical objects such as galaxies, stars and
  QSOs using a combination of labeled and unlabeled
  data.
• Our objectives were to demonstrate the
  usefulness of this semisupervised learning method
  and determine the effect of data set parameters
  (number of objects, number of features) on the
  classification accuracy of algorithm.

Applications Classify different galaxy
types into several distinct classes. Automatical
ly identify and group different types of
astronomical object into distinct classes. Use
cheap and abundant unlabeled data to improve
classification accuracy. Discover and isolate
new and interesting object classes for further
scientific analysis. Potential applications to
spacecraft health and safety monitoring and
spacecraft autonomy.
NASA/AISRP 2003-2006