From July 1 (2024) and for the next three months I will be closely working* with a Physics undergrad, named Simos Vlassis, who followed my Python course and wanted to dive into more interesting stuff with Python. Therefore, I found the opportunity to offer him the possibility to work on missing values techniques. This is important as in most astronomical problems missing values is the norm rather than the exception.
One particular case is the problem that Kostas Antoniadis is facing when trying to fit the Spectral Energy Distributions of Red Supergiants and in many cases some data are missing. More specifically the 24 μm band of Spitzer typically suffers from poor spatial resolution and consequently contamination that deems the photometry in this band useless. However, this point is very important to fit and understand the dusty component of these stars. There is a need to solve somehow this problem.
The approach in this case will be to explore various missing values imputation methods. Starting from the most basic ones (to establish a baseline model) we are going to explore as many as possible techniques exist and test them to see which one performs best. Depending on the progress we may also attempt to apply some (basic) machine/deep-learning regression models.
*This is a co-supervised internship with Andreas Zezas, performed at the premises of the Institute for Astrophysics, FORTH.