Inputs are very first handed by way of some totally related layer, into a double-layer residual multihead interest as shown in Fig. 7. Residual networks (Kaiming He, 2016), integrate feedforward to prevent neurons from suffering from exploding or vanishing gradients in the course of the training method. The absolutely linked levels inside the resid