ReLU do not becoming gradient 0 when input becomes larger which can prevent the Vanishing Gradient problem.