Machine learning is a topic that has risen in prominence recently as we get more and more data. We are seeing techniques from machine learning used more widely in astronomy. The goal of this reading group is to become more familiar with topics in machine learning and its connections to statistical tools that are in use in Astronomy. The plan is to go through a couple of textbooks on machine learning and discuss the basic underlying principles and methods. Each week, members of the reading group would present a topic with associated code implementing the algorithm.
The reading group github page is at: https://github.com/UCLAMLRG. Some of the links to books and resources are to the right.
Machine learning is well suited for classification problems where there are a lot of training examples. One of the applications I'm working on is training neural nets to recognize planet transits in light curves from the Kepler mission and applying it to K2 and TESS planet missions. I'm interested in how the use of convolution neural networks and generative adversarial networks for this task as well as evaluating how well they perform.
Determining whether a source in an image is a star can be non-trivial in many astronomical cases. For example, in an image crowded with many stars, such as at the Galactic center, stars can have a large range of brightnesses and may overlap each other in projection. In addition, adaptive optics imaging can cause variations in what a point source looks like (the point-spread function) across an image. I am am involved developing methods to detect and characterize the properties of stars and other objects in images. I am also interested in deep learning and Bayesian object detection to separate and identify stars.
Point-spread function reconstruction for integral-field spectrograph data, Do, Tuan; Ciurlo, Anna; Witzel, Gunther; Lu, Jessica; Turri, Paolo; Fitzgerald, Michael; Campbell, Randy; Lyke, Jim; Ghez, Andrea, 2018, Proceedings of the SPIE, Volume 10703