Stein Variational Gradient Descent

Description

Implementation of Stein Variational Gradient Descent with TensorFlow 2.0

Requirements

• tensorflow>=2.0.0-rc0
• tensorflow-probability>=0.8.0-rc0

Contents

• gaussian_mixture.py - Transformation initial samples to Gaussian mixture using SVGD.
• bayesian_logistic_regression.py - Bayesian Logistic Regression with SVGD.
• tfp_bayesian_logistic_regression.py - Bayesian Logistic Regression with Mean-Field Variational Inference.

Reference

• Q. Liu, and D. Wang, "Stein Variational Gradient Descent: A General Purpose Bayesian Inference Algorithm", in NIPS, 2017. https://papers.nips.cc/paper/6338-stein-variational-gradient-descent-a-general-purpose-bayesian-inference-algorithm