Justification of learning abstract math for machine learning ?

I wonder which research areas in ML requires a good amount of abstract math knowledge ?

I mean for manifold learning you need to understand topology, or for kernels you need functional analysis are there other areas you know with such reqs ?