Let's say I'm using a neural net to solve some problem where the output of the net is a vector of some size. What are the advantages (and disadvantages if there are any) of training a single net to output the whole vector versus training multiple nets, one to output each element in the vector?
$\begingroup$
$\endgroup$
