Active Learning For Identifying Function Threshold Boundaries (2005)

Brent Bryan, Jeff Schneider, Robert C. Nichol, Christopher J. Miller, Christopher R. Genovese, Larry Wasserman

Tags

Active Learning, Applications, Astrostatistics, Gaussian Processes, Statistical Data Mining for Astrophysics

Abstract

We present an efficient algorithm to actively select queries for learning
 the boundaries separating a
function domain into regions where the function is above and below a given
threshold.  We develop experiment selection methods based on entropy,
misclassification rate, variance, and their combinations, and show how
they perform on a number of data sets.  We then show how these algorithms
are used to determine simultaneously valid $1-\alpha$ confidence
intervals for seven cosmological parameters.  Experimentation shows
that the algorithm reduces the computation necessary for the parameter
estimation problem by an order of magnitude.

Full text

Download (application/pdf, 7.9 MB)

Approximate BibTeX Entry

@inproceedings{bryan:nips2005,
    Year = {2005},
    Journal = {NIPS Conference Proceedings},
    Organization = {NIPS},
    Author = { Brent Bryan, Jeff Schneider, Robert C. Nichol, Christopher J. Miller, Christopher R. Genovese, Larry Wasserman },
    Title = {Active Learning For Identifying Function Threshold Boundaries}
}