Dr. Andrej Prsa
Villanova University

University of Florida Astronomy Colloquium - Mar. 7th, 2007

Artificial Intelligence Approaches to Modeling Eclipsing Binary Stars

Eclipsing binary stars are among the most important celestial objects in astrophysics because they provide a unique way to determine absolute parameters of stars (their masses, radii, temperatures) without bias. Since eclipsing binaries can host any type of stars, be it main sequence, instability strip, giants or dwarfs, pulsars or black holes, they are essentially astrophysical laboratories. Eclipsing binaries are also extremely important for measuring distances and calibrating standard candles. Obtaining geometrical and physical parameters of eclipsing binaries involves solving the inverse problem: finding a set of parameters that produce synthetic data curves that best match observations. A promised scientific yield inspired early computer programs such as the renowned WD code (Wilson & Devinney, 1971) that have been gaining on accuracy and complexity for the last 3 decades. Yet a dawn of a new era is upon us, where robotic telescopes and dedicated surveys are acquiring massive amounts of data and tools devised for manual reduction and analysis are no longer adequate. New approaches that can handle hundreds of thousands of data curves in a fully automatic way are thus needed. The EBAI project (Eclipsing Binaries with Artificial Intelligence, http://www.eclipsingbinaries.org) explores one such approach - the use of back-propagating neural networks to determine principal parameters of eclipsing binaries. We present the scheme, benchmarks and the results of the EBAI approach.