4
$\begingroup$

Should VAEs be even used for non-generative tasks? If I were to use both models for embedding images, how would the embedding space differ on a structural level?

$\endgroup$

1 Answer 1

2
$\begingroup$

For the vanilla autoencoder the structure is like this:

enter image description here

It can be treated as a nonlinear extension of PCA, while for the variational autoencoder a mean and a standard deviation is added as a layer for each hidden variable in the middle layer:

enter image description here

For the detailed difference please refer to this answer.

Should VAEs be even used for non-generative tasks?

Yes, you can. The additional KL divergence(between variational distribution and the prior distribution/normal distribution) loss can be just seen a regularization and regularization can reduce the variance(at the risk of increasing bias).

If I were to use both models for embedding images, how would the embedding space differ on a structural level?

For the VAE the values of embedding would just be like samples from the normal distribution while that doesn't hold for the general autoencoder.

Reference:
Intuitively Understanding Variational Autoencoders

$\endgroup$
3
  • 1
    $\begingroup$ Sorry, but this only states what they are. I'm asking for how they differ when it comes to non-generative tasks such as embeddings. $\endgroup$
    – Daniel
    Aug 11, 2019 at 22:43
  • 1
    $\begingroup$ @Daniel OK. I will update that later. $\endgroup$ Aug 11, 2019 at 22:47
  • $\begingroup$ @Daniel Have I answered your question? $\endgroup$ Apr 20, 2020 at 14:59

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.