more wavelets

I was taken away from the utopia that is exoSAMSI to a meeting in San Francisco with the Moore and Sloan Foundations about data science. On the flight, I worked on writing up what we learned yesterday about wavelets. The main thing we learned—and I have to credit this to Baines (Davis) and Dawson (CfA)—is that even in the dumbest wavelet basis, the wavelet amplitudes for a detrended G-type star lightcurve look very close to indpendent, Gaussian draws. This fully justifies our wavelet-based simple Gaussian Process approach. What do I mean by detrended? Not saying, just yet, because we don't yet really know what works best. The Moore and Sloan meeting was fun; it involved a lot of brainstorming.

No comments:

Post a Comment