Invited Talk - Splinter JungeAG
Monday, 13 September 2021, 16:30 (virtual JAG)
Photometric Search for Exomoons by using Convolutional Neural Networks
Municipal Gymnasium Thomaeum
Context: To find exomoon candidates and to model these candidates light curves, algorithms are used that need much computational power per candidate and need to be fitted to each light curve. Aims: It is shown that exomoon signatures can be found by using deep learning and Convolutional Neural Networks (CNNs), respecetively, trained with synthetic light curves combined with real light curves with no transits. Methods: First, synthetic light curves with varying parameter-combinations were simulated by using a High-Performance-Computing Cluster. This data was combined with Kepler light curves to make the CNN distinguish between noise and actual signatures. By using this data, a well-fitting network architecture as well as a good preprocessing for the data to enable the network to find this weak signatures were developed and the best fitting network was trained with a bigger data set. Results: It is found that CNNs trained by combined synthetic and observed light curves may be used to find moons bigger or equal than roughly 2-3 earth radii. Conclusions: Using neural networks in future missions like Planetary Transits and Oscillation of stars (PLATO) might increase the speed of finding a exomoon candidate.