I am reading some research papers about graph convolutional neural networks and I have seen the term "latent representations" used a lot. For instance, "the model was able to learn latent representations of the nodes of the graph".

What does the word "latent" means here? Is it the same as latent features?


1 Answer 1


The latent representation is the simplified model of your input data, for example, created by a neural network.

Considering an autoencoder, the central layer of this network (after training) will contain a simplified representation of the input data (i.e. summary of key features), which can be used to reconstruct the output.

If we take a dictionary definition of Latent: present and capable of emerging or developing but not now visible, obvious, active, or symptomatic, we can see how this describes the somewhat non-existence of the state, rather instead only a latent representation of the input data.

This image is a nice description. The latent representation is key features of the input data (here: the ears, nose, eyes of the animals.)

enter image description here

So yes, the latent representation is the sum of the latent features.

  • 1
    $\begingroup$ I think you should reconsider the point i.e. " created by your neural network". It can be without a Neural Network. $\endgroup$
    – 10xAI
    Commented Mar 2, 2021 at 14:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.