Machine learning on cosmological simulations

Using the EAGLE simulations for a better ground truth

Supervising master students

Going from observations to physical properties (e.g. estimating the stellar mass) is very challenging

• Observations have uncertainties
• Our physical models have uncertainties and biases
• Different wavelength regimes can suffer from different biases

It is easier to forward-model simulated galaxies into observations

Via radiative transfer, we can create a mock image

Top right inset is a SKIRT image of an EAGLE galaxy

These simulated galaxies can then be compared to actual observations

This is a galaxy simulated by the FIRE simulations

Using machine learning, we can learn the relation between physical properties and observed properties, using simulated galaxies

This relation can then be applied to real galaxies

Find out more in Pieter's thesis!