Seminars and Colloquia at ESO Santiago

The Galactic Center provides a unique environment to study the effects of General Relativity in the vicinity of a supermassive Black Hole. Stars orbiting the Black Hole SgrA* serve as test particles to measure deviations from Newtonian motion. The near-infrared beam combiner instrument GRAVITY at ESO’s VLT has enabled the measurement of the gravitational redshift and Schwarzschild precession of the star S2, one of the predicted relativistic effects on its orbital motion.
To measure even higher order deviations from Newtonian motion, in particular, to constrain thereby the spin of SgrA*, new stars need to be found that are orbiting the Black Hole even closer than S2.
Here, image reconstruction of GRAVITY’s high-resolution interferometric data is a powerful tool for finding faint yet undiscovered stars.
We present the revised image reconstruction method GRAVITY-RESOLVE (G^R), based on Bayesian Inference, which is particularly designed for Galactic Center observations with GRAVITY. We give an overview of its development, most recent results, and its application to latest data of GRAVITY+, the upgrade of GRAVITY. In particular, we present the orbit of the star S301, a star that has been discovered with G^R, and its implication on the possibility of constraining the spin of SgrA* in the era of the ELT.

TBD

The exponential growth of astrophysical datasets from modern observatories and sky surveys demands robust, scalable, and intelligent analysis techniques. In this talk, I will present two distinct yet complementary applications of machine learning to major challenges in observational astrophysics: spectral line fitting and photometric redshift refinement.
First, I will introduce FLAME, a deep convolutional neural network framework designed to fit Voigt profiles to H I Lyman-α absorption lines. Trained on millions of simulated spectra modeled to mimic HST-COS observations, FLAME combines classification and regression networks to determine component structures and derive key physical parameters such as Doppler widths and column densities. It achieves high accuracy on simulated data, with performance degradation on real observations attributable to noise and instrumental effects, highlighting both the promise and challenges of applying supervised learning to spectroscopic data.
Second, I will demonstrate the use of self-organizing maps (SOMs) to assess and refine photometric redshift (photo-z) catalogs for bright galaxies in the Kilo-Degree Survey (KiDS). By mapping galaxy colors and magnitudes onto a low-dimensional manifold, we identify regions of parameter space where empirical photo-z estimates are less reliable due to underrepresentation in the spectroscopic training sets. This unsupervised approach enables targeted refinement of the catalog and guides the optimal incorporation of additional spectroscopic data.
Together, these case studies illustrate the power and versatility of machine learning--both supervised and unsupervised--in extracting physical insights from complex, high-dimensional astrophysical data. As data volumes continue to grow with upcoming missions like Rubin-LSST, Euclid, and DESI, such techniques will be essential for maximizing scientific return across domains.

TBD

TBD

tbd

TBD

TBD

TBD

TBD

TBD

TBD

TBD

TBD

TBD

TBD

TBD

tbd