Alp Kucukelbir, Rajesh Ranganath, Andrew Gelman, David M. Blei
TL;DR本文介绍了一种名为自动微分变分推断(ADVI)的方法,可以通过输入一个贝叶斯模型和数据集,自动确定适当的变分族并优化变分目标,而且支持广泛类别的模型,本方法在 Stan 中实现,可用于各类模型的变分推断。
Abstract
variational inference is a scalable technique for approximate Bayesian
inference. Deriving variational inference algorithms requires tedious
model-specific calculations; this makes it difficult to automate. We pr