Fig. 2:

Different layers of a sample autoencoder and the outputs. The input to the autoencoder is X and the output (reconstructed input) is $\widehat{\mathbf{X}}$. Encoder layers E₁ and E₂ the corresponding outputs are Z₁, Z₂. The output of the hidden layer H is Z. There are two decoder layers D₂ and D₁ with corresponding outputs Z′₂ and Z′₁ and respectively.