The implementation is not variational

[amir-abdi]

Hi Shengija,

Thank you for your work. The implementation does not seem to be a variational approach. The mapping between the x and z is deterministic with no Gaussian density estimation and sampling. I may be misinterpretting here, but I can't match this to the InfoVAE paper: https://arxiv.org/pdf/1706.02262.pdf

Thanks, Amir

[rojinsafavi]

Any updates on this?

[amir-abdi]

@rojinsafavi I implemented another library for disentanglement which also includes the infoVAE. https://github.com/amir-abdi/disentanglement-pytorch Check it out and let me know if it helped.

[jwparks]

any updates?

[Sean0719]

Who can give me some suggestions that how to get the Figure 1 in paper. Thanks.

[ecfm]

any updates?

[ShengjiaZhao]

Hi sorry for this late reply. I haven't been up to date on looking at this repo. In the tutorial, we use a simplified implementation without adding random noise. For the version with random noise please see this file https://github.com/ShengjiaZhao/MMD-Variational-Autoencoder/blob/master/mmd_vae_eval.py

[ecfm]

Thank you Shengjia!

On Wed, Jul 21, 2021 at 1:33 AM Shengjia Zhao @.***> wrote:

Hi sorry for this late reply. I haven't been up to date on looking at this repo. In the tutorial, we use a simplified implementation without adding random noise. For the version with random noise please see this file

https://github.com/ShengjiaZhao/MMD-Variational-Autoencoder/blob/master/mmd_vae_eval.py

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/ShengjiaZhao/MMD-Variational-Autoencoder/issues/4#issuecomment-883903394, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABMKI5AYNUAWAYRTEDJOU53TYZLZJANCNFSM4IGUSQ7Q .