The study of exoplanets --i.e. planets orbiting stars other than the Sun-- is revolutionising our knowledge about the physical processes behind planetary formation and evolution. In addition, recent detections bring us closer to the question about the existence of life beyond the solar system. Future instrumentation already under construction will allow us to tackle this and other exciting questions. However, most of the techniques employed for the detection and characterisation of exoplanets experience limitations related to the intrinsic variability of the host stars. Our vision of the smallest exoplanets --those of rocky composition, similar to our planet-- is often blurred by these effects. The exoplanet community is taking important steps to overcome these limitations; instrumental developments, the use of new observational techniques, and the improving of data analysis methods are all on the agenda.

In this talk, after a short overview of the current knowledge of exoplanets, I will introduce some of the limitations the field is currently facing. Then, I will discuss some of the efforts to produce and/or implement sophisticated data analysis techniques and machine learning models to overcome them. In particular, I will discuss work done by our team, including finite mixture models, Bayesian model comparison, and machine learning models to optimise instrumental performance and operation. I will focus on the use of Convolutional Neural Networks to detect exoplanets in time series of stellar radial velocity measurements. I will show that this method outperforms classical analysis techniques, and will outline the possibilities for extending it and improving it. With these and upcoming algorithms, we aim at bringing the smallest and lightest exoplanets in the Galaxy into focus.