Neural-networks

Why neural networks?

Neural networks are great because they do the feature engineering for us. They use backpropagation to learn the features that contribute the most to the data and form a strong hypothesis without us having to do feature engineering or think of how the data is modeled beforehand.

Deep hidden layers implicitly learn features that humans would take forever to figure out.

Neural network approximation theory

We use neural networks because of how their nonlinear activation functions in each layer allow us to approximate any function possible, not just linear functions.

NOTE

A non-linear activation function is essential for a neural network to work. Otherwise, neural networks can only learn linear patterns.