Fig. 2From: Unsupervised network embeddings with node identity awarenessDenoising Autoencoder. A corrupted instance \(\tilde {x}\) of a graph x is fitted into the Autoencoder’s input. The Autoencoder is trained to recover a cleaned version of the input by compressing it through a non-linear mapping fθ and mapping it back (through \(g_{\theta ^{\prime }}\)) to a reconstructed version of the original input graph xBack to article page