Skip-graph model

Artificial neural networks contain one or several hidden layers. Usually, the values of the weights of the hidden layers learned by the network are not particularly interesting: they are just parameters of the model and we only care about the model output. For instance, when using a CNN to predict whether your image is a dog or a cat, you just want to get the prediction out of your model. Embedding is the exact opposite; we are not directly interested in the model output, but in the weights learned by the hidden layer, which are the embedded words.