Artificial Intelligence
The Leaky ReLU Function and Its Role in Neural Networks
Leaky ReLU is an activation function derived from ReLU that replaces the zero output for negative inputs with a small slope, typically 0.01. This keeps the gradient from ever reaching zero, which prevents the dying neuron problem and stabilizes training in very deep convolutional, recurrent, and GAN networks.