I was wondering, triggered by this s4tf (current) restriction on using control flows, if anyone could answer the question, if a neural network could ‘learn’ such control flows? We know networks can learn boolean operators, when non-linear.
Background for the question is, it is often said one way to see Deep Learning is, it is an Implementation of the vision that Computers can learn from examples instead of humans telling them what to do in control flows. So, I wonder how far that vision could go: can a network learn such control flows in some way or do we always need to add them?
Any thoughts or pointers anyone?
@rxwei looping you, since this bug is assigned to you