“…II-C). (DNNs) [12,13], different types of AutoEncoders (AEs) [1,[33][34][35][36], one-and two-dimensional Convolutional Neural Networks (1D-and 2D-CNNs) [1,11,14,26,29,30,32,35,[37][38][39][40][41][42][43], variants of Recurrent Neural Networks (RNNs) such as Long Short-Term Memory (LSTM) [1,29,34,39,44,45] and Gated Recurrent Unit (GRU) [11,30,38,45,46], possibly exploiting the composition capabilities of hybrid DL architectures [1,30,44]. The way input data are fed to such architectures is paramount for taking full advantage of the DL paradigm.…”