Document Type

Honors Project On-Campus Access Only


Several fundamental experiments in physics and astronomy, such as the discovery of dark energy and the accelerated expansion of the Universe, have relied on the study of transient events -- short-lived astrophysical sources which allow only a narrow time window for study before fading into obscurity. Transient studies have largely relied on humans with significant domain expertise to identify these rare events within the thousands of detections, and tens of thousands of false detection artifacts from wide-field astronomical surveys. However, with volumes of data from astronomical surveys increasing rapidly, moving past human inspection of these transient phenomena has now become of paramount importance. In this senior honors thesis, I outline an automated method to analyze, characterize and classify transient events discovered by large-scale synoptic surveys. This project combines algorithms from signal processing and machine-learning, and applies them to a complex astrophysical problem. I explore the properties of a specific class of transient phenomenon: supernovae. I attempt to restructure the sparse, unevenly sampled, heteroskedastic data from various existing observational campaigns. I use Gaussian Processes to generate a non-parametric representation of the time-domain supernovae data, or "light curves'", apply wavelet methods for feature extraction that allow for approximate translation invariance, and employ a Random Forest classifier algorithm to distinguish between supernovae types. Initial results from our classification scheme indicate good performance for all wavelet classes. The classification code will be used in a stage of the ANTARES pipeline, created for use on the upcoming Large Synoptic Survey Telescope and other precursor wide-field surveys.



© Copyright is owned by author of this document