Eva van Weenen used neural networks on the EAGLE simulations to connect galaxy spectra with stellar mass

Eva van Weenen, master student at Leiden University

I have had the pleasure to be Eva van Weenen‘s main supervisor for her master research project at Leiden University. Eva has recently completed the Master degree in Data Science & Astronomy. For her thesis, she built artificial neural networks to study the relationship between modelled ugriz broad-band fluxes of the Sloan Digital Sky Survey (SDSS) and the stellar mass, determined using galaxies from the EAGLE (Evolution and Assembly of GaLaxies and their Environment) RefL0100N1504 simulation at redshift z ∼ 0.1 (Schaye et al., 2015; Trayford et al., 2015).

The hyperparameters of Eva’s neural network were optimised using Tree-structured Parzen Estimators with 5-fold cross-validation. Eva trained the network on the fluxes-stellar mass relation from EAGLE, and later applied it to SDSS data with stellar masses determined by Chang et. al (2015). She obtained an accuracy of R^2 = 0.99 for EAGLE galaxies, and R^2 = 0.94 for SDSS galaxies, indicating a discrepancy in the photometry-stellar mass relation of eagle and SDSS galaxies.

Eva’s work will be publish soon. If you are interested in the neural network, more information can be found in Eva’s github page:

Congratulations Eva!