Search

The field of exoplanets has enjoyed unprecedented growth in the past decades, planets are being discovered at an exponential rate. With the launch of next-generation facilities in the coming decades, the arrival of high-quality spectroscopic data is expected to bring about yet another revolutionary change in our understanding of these remote worlds. The field has been actively developing tools to comprehend the large stream of incoming data, and among them, Machine Learning techniques are building up momentum as an alternative to conventional approaches. In this work, I developed methodologies to uncover potential biases in the interpretation of the exoplanetary atmosphere introduced during data analysis. I showed that naively combining observations from different instruments might lead to biased results, and in some extreme cases like WASP-96 b, it is impossible to com- bine observations. A new scheme of retrieval framework, namely the L - retrieval, holds the potential to detect incompatibility among different datasets by combining light-curve fitting with atmospheric radiative transfer modelling. This work also documents the application of ML techniques to two distinct fields of exoplanetary science: a planet signal detection pipeline for direct imaging data and a suite of diagnostic tools designed for the characterisation of exoplanets. In both approaches, I pioneered the integration of Explainable AI techniques to improve the reliability of the deep learning models. Initial successes of these novel methodologies have provided an exciting prospect to tackle upcoming challenges with the use of Artificial Intelligence. How- ever, significant work remains to progress these models from their current proof-of- concept stage to general application framework. In this thesis, I will discuss their current limitations, potential future, and the next steps required.