Categories

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.