In contrast, with a non-linear activation function we can increase the distance between A1 and A2 while we decrease the distance between B1 and B2. We can make B1 close to A1, but B2 distant from A1. By applying non-linear functions, we create new relationships between the points. With every new non-linear transformation we can increase the complexity of the relationships. In deep learning, using non-linear activation functions creates increasingly complex features with every layer.