Machine learning algorithms are very good at approximating very general functions. Sometimes we want to constrain some parameters (e.g. cosmological parameters) given a set of observations. For standard observables (e.g. galaxy clustering) the likelihood function behind the data is known. By sampling it, constraints on the parameters can be placed.

​

It is also possible to use neural networks to approximate the function that relates an observable (with unknown likelihood) to the value of the cosmological parameters. 

​

We have used this method to constrain the value of the cosmological and astrophysical parameters from measurements of the star-formation rate density (SFRD) of the SIMBA simulations from z=0 to z=7.