Tanh-like Activation Function Implementation for High-performance Digital Neural Systems
S. Marra, M.A. Iachino, Francesco Carlo Morabito · 2006
In this paper, a high-performance implementation of a programmable tanh-like activation function is presented. The proposed function can be successfully used to train neural networks obtaining generalization ability slightly better than standard activation functions. The achieved accuracy, the high computational speed and the small amount of area resources make the proposed solution ideal for digital implementations.