Gaussian Pre-Activations in Neural Networks: Myth or Reality?