Gaussian Mixture model VAE version for keras

Henry Miller

unread,

Jul 20, 2021, 10:54:40 PM7/20/21

to Keras-users

Hi, does anyone know if there is any GMM version of VAE model in keras, preferably a notebook version? What I am looking for is something similar to this:

https://blog.keras.io/building-autoencoders-in-keras.html

What I am trying to do is to replace the standard prior normal with a Gaussian Mixture but I am not sure how to modify that code to acommodate this. I have looked at other reference codes but there are not robust enough to generate arbitrary datasets. Thanks!!

Sayak Paul

unread,

Jul 20, 2021, 11:03:53 PM7/20/21

to Henry Miller, Keras-users

There's an official VAE tutorial: https://keras.io/examples/generative/vae.

Sayak Paul | sayak.dev

--
You received this message because you are subscribed to the Google Groups "Keras-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to keras-users...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/keras-users/55ab0efe-b45a-4082-a9d6-65a04e0bb90cn%40googlegroups.com.

Henry Miller

unread,

Jul 22, 2021, 4:08:10 AM7/22/21

to Keras-users

Hi,

Thanks for the reply. I have looked that website and those are just the default examples. What I mean is, is the a GMM equivalent for vae, something that is more advanced, not that VAE in that page because those uses the standard normal prior.

Sayak Paul

unread,

Jul 22, 2021, 4:47:36 AM7/22/21

to Henry Miller, Keras-users

If you can lay out a few more details I think we can work together on an example. Which bit you would change in the existing tutorial, what you formulate in place of those, etc.

Sayak Paul | sayak.dev

On Thu, Jul 22, 2021 at 1:38 PM Henry Miller <henrymiller....@gmail.com> wrote:

Hi,

Thanks for the reply. I have looked that website and those are just the default examples. What I mean is, is the a GMM equivalent for vae, something that is more advanced, not that VAE in that page because those uses the standard normal prior.

On Wednesday, July 21, 2021 at 11:03:53 AM UTC+8 spsay...@gmail.com wrote:
There's an official VAE tutorial: https://keras.io/examples/generative/vae.
Sayak Paul | sayak.dev

On Wed, Jul 21, 2021 at 8:24 AM Henry Miller <henrymiller....@gmail.com> wrote:
Hi, does anyone know if there is any GMM version of VAE model in keras, preferably a notebook version? What I am looking for is something similar to this:

https://blog.keras.io/building-autoencoders-in-keras.html

What I am trying to do is to replace the standard prior normal with a Gaussian Mixture but I am not sure how to modify that code to acommodate this. I have looked at other reference codes but there are not robust enough to generate arbitrary datasets. Thanks!!

--
You received this message because you are subscribed to the Google Groups "Keras-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to keras-users...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/keras-users/55ab0efe-b45a-4082-a9d6-65a04e0bb90cn%40googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Keras-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to keras-users...@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/keras-users/482e9c6d-82d3-450b-b50d-7dc40729c311n%40googlegroups.com.

Henry Miller

unread,

Jul 22, 2021, 4:49:06 AM7/22/21

to Keras-users

https://github.com/QingyuZhao/TruncatedGaussianMixtureVAE/blob/master/mnist.ipynb something like what this guy is doing, but I dont understand what is the difference and what makes it a GMMVAE compared to the ordinary VAE

Henry Miller

unread,

Jul 22, 2021, 5:13:36 AM7/22/21

to Keras-users

refer to comment below

On Thursday, July 22, 2021 at 4:47:36 PM UTC+8 spsay...@gmail.com wrote:

Reply all

Reply to author

Forward