Leveraging the infinite dimensional neural network architecture we proposed
in arXiv:2109.13512v4 and which can process inputs from Fr\'echet spaces, and
using the universal approximation property shown therein, we now largely extend
the scope of this architecture by proving several un