The search and characterization of exoplanets is an ever-expanding field of astrophysics and with new instruments, like the James Webb Space Telescope, coming online in the next years, developing new and faster methods, like the one proposed here, to analyze and characterize the massive amount of exoplanets will be crucial to the advance of the field.
Atmosphere characterization of exoplanets presents a challenge, as the number of parameters to consider to produce an effective model of a planet’s atmosphere is tremendous. In the recent years, it has become quite clear that deep learning tools can be an effective way to model complicated systems, but these tools have yet to be integrated in the astronomer’s usual toolbox. My proposed research project is to develop a novel approach to atmospheric retrievals using deep learning architectures, such as neural networks, to produce accurate models in a shorter amount of time than traditional simulations.
Atmospheric characterization of exoplanets is an active area of research and thus a lot of different models exist and do an acceptable job at retrieving important features of a planet, such as its temperature profile or which molecules are present in its atmosphere. But these models are computationally expensive and thus producing many of them with different parameters to compare with observational results can take days if not weeks. Neural networks, on the other hand, can take a long time to train, but once they are, they usually output their results pretty fast. Neural networks or other deep learning tools could be trained on accurate, but computationally expensive, atmospheric models of exoplanets in the hope of replicating the results of these retrievals, but in less time than needed to run full atmospheric retrievals.