How is a neural network able to model non-linear relations when it uses linear combinations of the inputs? Won’t we simply get a linear regression model? I have read people claim Neural Network are powerful enough to be able to correctly predict the output without knowing the form of the relation between input and output, i.e if its linear or polynomial

1 Like

Hi!

This is explained in Week 2 Video “Why do we need activation functions?” and

further expanded on in the ReLU activation Lab

It works because the hidden layer always uses a non-linear activation function.

thanks, I was at week1 while I had this doubt.

A very good doubt indeed.