Pre-study of transit recognition with machine learning
The past and current space-based transit surveys (Kepler/K2 and TESS) are goldmines for our knowledge of extrasolar systems and their architecture (spatial and mass repartition of exoplanets). However, these mines haven’t been exploited to their full potential yet. As shown in recent papers co-authored by Kepler science-team members, small (e.g. terrestrial) planets whose orbit is heavily perturbed by planet-planet interactions cannot be recovered by existing data-processing tools.
The idea of this project is to collaborate with machine learning experts working in the company Disaitek to develop different algorithms to detect new planets that cannot be recovered by existing current data-processing tools.
Market opportunities and industrial applications
The company acquires expertise in the latest time series analysis methods that would naturally transfer to some of their other activities, including data pre-processing algorithms that are specific to the astrophysics field. Moreover, the company will benefit from the experience of working with large scale datasets of real data and a well-defined physical problem.
Contacts
PlanetS partners: Adrien Leleu, Yann Alibert and Stéphane Udry.
Non-PlanetS partner: Châtel Grégory, Disaitek (webpage and email).
Publications and Patents
In case of detection of exoplanets with this method, possible peer-reviewed publications are foreseen.
