In comparison to Sigmoid, ReLU based functions train much faster, are more expressive than logistic function, prevent gradient vanishing.