I'm not sure that's actually correct. In fact I'm sure it is incorrect for a polynomial of degree 1. Otherwise, there's nothing special about relu or tanh that you can't use sequential/backprop on polynomial regression in general.