Kernel based method applied on irregularly sampled time series

Cuevas-Tello, J.C., Tino, P. and Raychaudhury, S.
The Analysis of Patterns Workshop (poster) Erice, Italy (Ettore Majorana Centre for Scientific Culture). Directors: N. Cristianini, R. Cerulli and J. Shawe-Taylor., 2005.


Abstract: Due to gravitational lensing, an astronomer often observes multiple light images of the same distant celestial object (e.g. quasar). Since the images correspond to different light travel paths, the variation of light observed in these images are not synchronous. In practice, if a lensed source is monitored by a telescope, the light curves that would be observed as irregularly sampled time series (with measurement errors) from each image would have a delay between them. Our aim is to find the time delay between a pair of such time series. We present a novel kernel-based method for estimating the time shift in light curves due to gravitational lensing. We compare our approach with several widely used statistical techniques on a large set of artificially generated light curve data, as well as on real data (quasar 0957+561).