Learning to Draw Samples with Amortized Stein Variational Gradient Descent

Yihao Feng, Dilin Wang, Qiang Liu
Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence, PMLR R15:521-530, 2017.

Abstract

We propose a simple algorithm to train stochastic neural networks to draw samples from given target distributions for probabilistic inference. Our method is based on iteratively adjusting the neural network parameters so that the output changes along a Stein variational gradient direction (Liu & Wang, 2016) that maximally decreases the KL divergence with the target distribution. Our method works for any target distribution specified by their unnormalized density function, and can train any black-box architectures that are differentiable in terms of the parameters we want to adapt. We demonstrate our method with a number of applications, including variational autoencoder (VAE) with expressive encoders to model complex latent space structures, and hyper-parameter learning of MCMC samplers that allows Bayesian inference to adaptively improve itself when seeing more data.

Cite this Paper

BibTeX
@InProceedings{pmlr-vR15-feng17a, title = {Learning to Draw Samples with Amortized {S}tein Variational Gradient Descent}, author = {Feng, Yihao and Wang, Dilin and Liu, Qiang}, booktitle = {Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence}, pages = {521--530}, year = {2017}, editor = {Elidan, Gal and Kersting, Kristian}, volume = {R15}, series = {Proceedings of Machine Learning Research}, month = {11--15 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/r15/main/assets/feng17a/feng17a.pdf}, url = {https://proceedings.mlr.press/r15/feng17a.html}, abstract = {We propose a simple algorithm to train stochastic neural networks to draw samples from given target distributions for probabilistic inference. Our method is based on iteratively adjusting the neural network parameters so that the output changes along a Stein variational gradient direction (Liu & Wang, 2016) that maximally decreases the KL divergence with the target distribution. Our method works for any target distribution specified by their unnormalized density function, and can train any black-box architectures that are differentiable in terms of the parameters we want to adapt. We demonstrate our method with a number of applications, including variational autoencoder (VAE) with expressive encoders to model complex latent space structures, and hyper-parameter learning of MCMC samplers that allows Bayesian inference to adaptively improve itself when seeing more data.}, note = {Reissued by PMLR on 04 October 2026.} }
Endnote
%0 Conference Paper %T Learning to Draw Samples with Amortized Stein Variational Gradient Descent %A Yihao Feng %A Dilin Wang %A Qiang Liu %B Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2017 %E Gal Elidan %E Kristian Kersting %F pmlr-vR15-feng17a %I PMLR %P 521--530 %U https://proceedings.mlr.press/r15/feng17a.html %V R15 %X We propose a simple algorithm to train stochastic neural networks to draw samples from given target distributions for probabilistic inference. Our method is based on iteratively adjusting the neural network parameters so that the output changes along a Stein variational gradient direction (Liu & Wang, 2016) that maximally decreases the KL divergence with the target distribution. Our method works for any target distribution specified by their unnormalized density function, and can train any black-box architectures that are differentiable in terms of the parameters we want to adapt. We demonstrate our method with a number of applications, including variational autoencoder (VAE) with expressive encoders to model complex latent space structures, and hyper-parameter learning of MCMC samplers that allows Bayesian inference to adaptively improve itself when seeing more data. %Z Reissued by PMLR on 04 October 2026.
APA
Feng, Y., Wang, D. & Liu, Q.. (2017). Learning to Draw Samples with Amortized Stein Variational Gradient Descent. Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research R15:521-530 Available from https://proceedings.mlr.press/r15/feng17a.html. Reissued by PMLR on 04 October 2026.

Related Material