thjashin / spectral-stein-grad

Code for "A Spectral Approach to Gradient Estimation for Implicit Distributions" (ICML'18)

spectral-stein-grad

Update: Please find our latest python library for nonparametric score estimation (including the spectral estimator) at https://github.com/miskcoo/kscore

---

This repository contains the official code for

A Spectral Approach to Gradient Estimation for Implicit Distributions (ICML 18)

https://arxiv.org/abs/1806.02925

Dependencies

• Tensorflow >= 1.8
• ZhuSuan >= 0.3.1

Get Started

python -m toy.guassian

Implicit VAEs

Train plain VAE on CelebA:

python -m vae.vae_celeba

Train implicit VAE on CelebA using entropy gradients estimated by SSGE:

python -m vae.vae_celeba_implicit

Citation

To cite this work, please use

@InProceedings{shi2018spectral,
  title = 	 {A Spectral Approach to Gradient Estimation for Implicit Distributions},
  author = 	 {Shi, Jiaxin and Sun, Shengyang and Zhu, Jun},
  booktitle = 	 {Proceedings of the 35th International Conference on Machine Learning},
  pages = 	 {4651--4660},
  year = 	 {2018},
}

spectral-stein-grad's Issues

codes for the other two experiments?

Hi, I find in this repository there are only codes for two experiments in the ICML paper. Could you release the code for HMC and BNN experiments?

thanks!