ReCA: A Parametric ReLU Composite Activation Function

John Chidiac, Danielle Azar

Activation functions have been shown to affect the performance of deep neural networks significantly. While the Rectified Linear Unit (ReLU) remains the dominant choice in practice…