It is a pretty cool result. Basically the lay summary (from my understanding) is that a multivariate function with n-arguments can be perfectly represented by a

It is a pretty cool result. Basically the lay summary (from my understanding) is that a multivariate function with n-arguments can be perfectly represented by a network with 2 hidden layers (one having width n and another width 2n) and two activation functions.