Finding the Odd One Out in Large Spectroscopic Surveys

Details Finding the Odd One Out in Large Spectroscopic Surveys Finding the Odd One Out in Large Spectroscopic Surveys

Jan 2011

adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2011aas...21715304c&link_type=abstract

American Astronomical Society, AAS Meeting #217, #153.04; Bulletin of the American Astronomical Society, Vol. 43, 2011

Statistics

Scientific paper

Through shear volume of data, next generation surveys will provide us with unprecedented and detailed information about the full distribution of astronomical sources. Anomalous events influenced by the the most extreme and interesting physics will no longer be relegated to the dustbin of "small number statistics." Unfortunately, that same volume of data will render the task of culling these extreme events from the background of ordinary stars and galaxies virtually impossible. Both the number of events and the dimensionality of the data (e.g. a spectral energy distribution measured in 4000 wavelength bins) exist well outside the reasonable limits of human processing In this context, we seek algorithms to project N>>1 dimensional data down to 2 or 3 effective dimensions, preserving the physics of the correlations within the unprojected data. Inspection in these effective dimensions then allows us to identify both objects that resemble one another (classification of objects) and objects that resemble nothing at all (anomaly detection). We consider both Principal Component Analysis, which attempts the projection under the assumption that any given data point can be reconstructed from a linear combination of all other data points, and Local Linear Embedding, which attempts to reconstruct data points only from their nearest neighbors, preserving the non-linear relationships between different neighborhoods. We use stellar spectra from the SDSS to show how these techniques can identify interesting classes of astronomical sources. We acknowledge support from the DOE Applied Mathematics Program DE-FG02-87ER40315

Affiliated with

Also associated with

No associations

Rating

Finding the Odd One Out in Large Spectroscopic Surveys does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.

If you have personal experience with Finding the Odd One Out in Large Spectroscopic Surveys, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Finding the Odd One Out in Large Spectroscopic Surveys will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFWR-SCP-O-1395952