Abstract

Contributed Talk - Splinter JungeAG

Monday, 12 September 2022, 14:40   (SFG 1010 / virtual JAG)

An Exploration of Machine Learning Algorithms for Photometric Detection of Exomoons

Lukas Weghs
Gymnasium Thomaeum

The search for exomoons is very computationally expensive if one uses e.g. nested sampling. In order to save resources, different teams of scientists employed and evaluated, respectively, Convolutional Neural Networks (CNNs) in the search for exomoons last year. But so far, no exomoon has been confirmed. These research projects employed CNNs trained with single transits. In contrast, I evaluate a different preprocessing method giving me a 2-dimensional representation of the light curve extracts based on five different transits instead of just one transit. Then, I aim to explore various machine learning methods and combine them by using Stacked Generalisation. Thereby, I can merge the advantages of these different ML-based algorithms.