Calendar

ABSTRACT: Atmospheric retrievals on exoplanets involve usually computationally intensive Bayesian methods. The choice of the fitting parameters bounds are often leaded by physical constraints and the user experience. In these paper we introduce an alternative method that can help to automatically define the boundary conditions of the model and set a reliable parameters space for a Bayesian analysis. We show how a new generation of neural networks, a Generative Adversarial Network (GAN), can learn how to reproduce a transmission spectrum and understand how it depends on the planetary physical parameters.