Deep Learning for Small and Noisy Astrophysical Datasets: From Cosmology to Exoplanet Detection

Deep learning is increasingly revolutionizing astrophysics by enabling advanced analysis of complex and high-dimensional datasets. However, the application of these methods is often limited by data availability and noise, requiring careful model design, optimization, and regularization to balance performance and robustness.

This talk presents recent research addressing these challenges, including deep learning techniques for cosmological parameter reconstruction, the use of genetic algorithms to optimize neural networks and improve precision, and the combination of these methods to accelerate Bayesian inference. It then focuses on applications to the detection of Earth-like exoplanets using stellar radial-velocity measurements, where stellar activity and instrumental effects dominate the signal. In particular, deep learning models trained directly on real high-resolution stellar spectra are shown to recover tiny Doppler shifts in unseen data, approaching the precision required for the detection of terrestrial planets.

The seminar aims to provide practical insights into the effective use of deep learning for astrophysical problems in which data are limited, and precision is crucial.

Fecha y lugar: 19/02/2026 – 12:30 | Salón de Actos
Isidro Goméz Vargas
IAA-CSIC