Galaxy formation is a highly non-linear process.  Gravitational interactions alone drive an initially Gaussian random field towards a highly non-Gaussian field with high peaks and large voids.

Observational effects add additional complexity so that the observed galaxy distribution is a non-Gaussian and inhomogeneous field.  Standard analyses approaches treat this field naively and miss much of the information encoded in the observations.   Machine learning enables rapid emulation of the galaxy formation process and provides a powerful tool for joint analyses of multiple surveys.

I will outline recent progress in this area.