Dustin Swarm

The amount of information available to astronomers through survey missions continues to grow exponentially, but much of that data will never be individually examined by a scientist. This information overload necessitates a Big Data approach to astronomy. Machine learning techniques are useful in helping astronomers search for specific targets in the deluge of data. Working with my advisor, Prof. Casey DeRoo at the University of Iowa, I adapted a random forest (RF) machine learning algorithm to look for unusual X-ray objects in the Chandra Source Catalog (CSC), a database of approximately 315,000 X-ray sources observed by NASA’s Chandra X-ray Observatory (CXO) over the past two decades. The RF categorizes the CSC objects by grouping together sources with similar physical properties. This enabled me to produce a catalog of CSC sources consistently identified as isolated outliers by the algorithm.

I will now turn my efforts to analyzing what makes these outliers unique from the rest of the sources in the CSC by utilizing data from existing CXO observations, such as spectra and emission-mechanism models, as well as proposing future follow up CXO observations. This work supports Objective 1.1 of the 2018 NASA Strategic Plan by leveraging the substantial data archive built by CXO to extend scientific discovery beyond individual observations. Looking for outlier sources within the CSC increases the odds of discovering sources that offer new insight into astrophysical phenomena or are in a relatively short-lived evolutionary state. Furthermore, our methodology serves as a blueprint for addressing the problem of data saturation in a manner that is easily applicable to upcoming large datasets, such as those of JWST and WFIRST.