BriefGPT.xyz
Jun, 2019
黑盒变分推断的可证明梯度方差保证
Provable Gradient Variance Guarantees for Black-Box Variational Inference
HTML
PDF
Justin Domke
TL;DR
本文通过研究在目标平滑、变分族为位置-尺度分布情况下的重参数估计器,为基于随机梯度估计器的最新变分推断方法提供了不可改善的界限。
Abstract
Recent
variational inference
methods use
stochastic gradient estimators
whose variance is not well understood. Theoretical guarantees for these estimators are important to understand when these methods will or wi
→