BriefGPT.xyz
Nov, 2020
关于深高斯过程变分推断中信噪比问题的研究
On Signal-to-Noise Ratio Issues in Variational Inference for Deep Gaussian Processes
HTML
PDF
Tim G. J. Rudner, Oscar Key, Yarin Gal, Tom Rainforth
TL;DR
通过理论分析和大量实验证明,在Deep Gaussian Processes (DGPs)使用加权变分推断训练的梯度估计中,信噪比问题是普遍存在的。本文提出的基于双重重参数化的梯度估计方法解决了这个问题,提高了DGPs模型的预测性能。
Abstract
We show that the
gradient estimates
used in training
deep gaussian processes
(DGPs) with
importance-weighted variational inference
are sus
→