Advanced Deep Learning with Python

Graph autoencoders

Let's have a quick recap of autoencoders, which we first introduced in Chapter 5, Generative Models. An autoencoder is an FFN that tries to reproduce its input (more accurately, it tries to learn an identity function, ). We can think of the autoencoder as a virtual composition of two components—the encoder, which maps the input data to the network's internal latent feature space (represented as vector z), and the decoder, which tries to reconstruct the input from the network's internal data representation. We can train the autoencoder in an unsupervised way by minimizing a loss function (known as a reconstruction error)