Department of Statistics and Data Science
Carnegie Mellon University
Incorporating Spectral Features into Classifying Astronomical Light Curves
Many inference challenges in cosmology can be placed into a common framework of three stages which taken together achieve the ultimate goal of estimating key cosmological parameters from the observables. After a brief overview of this framework, this talk will describe an important example of one such inference problem, that of classification of celestial objects from limited, noisy observations. Upcoming astronomical surveys will require such tools in order to meet scientific objectives. Typical observations of a candidate will consist of brightness measurements at irregularly-spaced time points, through each of a few filters. As is a recurring theme in cosmology, realistic methods will require one to incorporate current understanding of the physics behind the processes that generated the observable data. We describe a statistical classification scheme that combines (1) the physical knowledge of the relationship between the type of object and spectroscopic information – in a training set – with (2) the time series that are observed.
Friday, 4/20/2013, 11:30 AM, BLOC 113