Leaky ReLU is a variant of the ReLU activation function that allows a small, non-zero gradient for negative inputs. It mitigates the “dying ReLU” problem by preventing gradients from becoming zero when inputs are negative. This can lead to improved training stability in deep networks.