This project motivated by the STARE project of HAO, NCAR has two aspects. 

1. The detection of variable stars

Variable stars show variations (or periodicity) as a function of time in 
brightness. We develop the methods of estimating the period and the light 
curve of a variable star. A method for estimation of the period based on 
smoothing spline regression is suggested. This method, which finds the 
period to minimize GCV function, matches intensive visual searching method.
However, since GCV function is sensitive with outliers, we suggest quantile 
spline regression method as a robust method for finding the period. 
For estimating periodic function, well-studied nonparametric smoothing 
method such as smoothing spline and wavelet regression are used. 
The matter on multi-periodicity in variable stars is also considered.
We apply backfitting algorithm to find several periods from a variable star.

2. The classification of variable stars

Once we decide the light curves and periods of variable stars, we classify
stars into different groups according to amplitude, the range of period, 
the shape of light curve, and a lot of prior knowledge. For this problem, 
we will use traditional CART, Bayesian CART, and neural networks.