Characterization of Extrasolar Systems with Machine Learning

The existence of compact extrasolar systems of several super-Earth planets has recently been revealed by transit photometry, notably by the Kepler mission and the TRAPPIST telescope.
The transit detection method provides an insight into the physical and orbital parameters of the detected systems, but suffers from large observational uncertainties and unknowns.
Determining these unknown parameters remains extremely challenging and computationally expensive due to the large parameter space to explore. Fortunately, it is natural to assume that the detected extrasolar systems are long-term stable, otherwise their observation would be an extremely rare event. On the one hand, long-term stability requirements can be assessed with powerful tools of celestial mechanics, in particular chaos indicators. On the other hand, machine learning and deep learning techniques have the potential to go beyond classical long-term N-body approaches for predicting the stability of extrasolar systems and therefore guiding the search for realistic system parameters. In the CAML project, we aim to investigate the benefit of combining the celestial mechanics and machine learning disciplines to better characterize the detected extrasolar systems in computationally efficient ways. This project is especially timely given the recent launch of CHEOPS space telescope and the current preparation of the PLATO mission, giving us the opportunity to be fully operational before the next observations.