Skip to content

A modification of the original SVAE model with Hasting Metropolis sampling

License

Notifications You must be signed in to change notification settings

EmmaBouhanna/HM-SVAE

Repository files navigation

HM-SVAE

A modification of the original SVAE model with Hasting Metropolis sampling

About

This project is a variant of the S-VAE algorithm originally implemented in Davidson, Falorsi, De Cao et al., Hyperspherical Variational Auto-Encoders, which is available here : https://github.com/nicola-decao/s-vae-tf.

In this repository

  • hyperspherical_vae : original library from Davidson & al, which we modified a little bit to replace the sampling method in the latent space by Hasting Metropolis sampling.

  • Train_HM.ipynb : allows to train a model with 100 epochs in 2 dimension. Uses the mnist.py file. This notebook shows visualizations of training performances, and testing performances

  • illustration_vMF.ipynb : allows to visualize sampling of von Mises-Fisher distribution, both with the original sampling method and the Hasting metropolis sampling.

  • gaussian_to_hypersphere.ipynb : Small experiment to show the soap bubble effect.

  • reconstructing_MNIST.ipynb : Notebook representing figures for the SVAE latent space with HM sampler. Visualizations of MNIST reconstruction of latent space sampling.

Authors

Sebastian Partarrieu & Emma Bou Hanna

About

A modification of the original SVAE model with Hasting Metropolis sampling

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published