Skip to content
code for the paper "Stein Variational Gradient Descent (SVGD): A General Purpose Bayesian Inference Algorithm"
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
matlab Add comments Aug 18, 2016
python Add comments Aug 18, 2016
.Rapp.history vgd 1.0 Jul 18, 2016
LICENSE add license Nov 23, 2016
_config.yml Set theme jekyll-theme-cayman May 21, 2017

Stein Variational Gradient Descent (SVGD)

SVGD is a general purpose variational inference algorithm that forms a natural counterpart of gradient descent for optimization. SVGD iteratively transports a set of particles to match with the target distribution, by applying a form of functional gradient descent that minimizes the KL divergence.

The package contains implementation of SVGD in both Matlab and Python language. Demos are also provided to reproduce the results in our paper. The Bayesian neural network example is based on Theano.

For more information, please visit our project website - SVGD.

Toy example on 1D Gaussian Mixture

Toy example with 1D Gaussian mixture. The red dashed lines are the target density function and the solid green lines are the densities of the particles at different iterations of our algorithm (estimated using kernel density estimator).

Basic Usage

  x0: initial particles
  dlnprob: returns first order derivative of log probability
  n_iter: number of iterations
  stepsize: initial learning rate 
theta = SVGD().update(x0, dlnprob, n_iter, stepsize)


Qiang Liu and Dilin Wang. Stein Variational Gradient Descent (SVGD): A General Purpose Bayesian Inference Algorithm. NIPS, 2016.


Feedback is greatly appreciated. If you have any questions, comments, issues or anything else really, shoot me an email.

All rights reserved.

You can’t perform that action at this time.