Abstract:
Many astrophysical questions are addressed by large area surveys, where sensitivity limits the distance and luminosity of directly detected objects. One way to gain further effective sensitivity is to forego knowledge of individual objects and estimate the aggregate properties of many objects using a technique called stacking. With this approach, images of multiple objects taken from a survey map are used to create a single aggregate image. The typical approach to stacking is to use observed or derived properties in making sub-sample selections in order to seek trends. What has not been explored is designing algorithms to select the sample itself, based on goal-orientated stacking objectives, such as maximising or minimising the aggregate result from stacking. This thesis aims to design a suite of machine learning algorithms that will be employed to select object sub-samples from derived and measured quantities representing each object. This novel approach to statistical inference of radio sources may also have additional applications in statistical self-calibration of radio interferometric data, which would use the same algorithmic framework.
