Say I have 500 variables and I believe those variables can be shown in a 4-dimensional latent representation which I want to learn.

What I have for training is 100K samples, and those samples are coming mainly from 3 unbalanced groups: 1st group has 1K samples, 2nd group has 49K samples, and 3rd group has 50K samples.

Do you think I can learn a meaningful representation by training a (variational) autoencoder with this data? Is there a reason that requires all samples to come from the same distribution? If not, is there a reason that requires balanced classes?

  • $\begingroup$ Can you please create paragraphs in your question body? $\endgroup$ – quintumnia Jan 7 '18 at 12:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.