ReLU is an activation function f(x) = max(0, x). It is computationally efficient and helps alleviate the vanishing gradient problem.
Key 2010s innovation that enabled deeper networks.
The most widely used activation function in vision.