Marton Havasi, José Miguel Hernández Lobato, Juan José Murillo Fuentes
TL;DR本研究使用随机梯度哈密尔顿蒙特卡洛方法对深层高斯过程模型的非高斯后验分布抽样，提供了一种新的推断方法，成为 Deep Gaussian Processes 领域新的最优模型。
Abstract
deep gaussian processes (DGPs) are hierarchical generalizations of Gaussian Pro- cesses that combine well calibrated uncertainty estimates with the high flexibility of multilayer models. One of the biggest challenges with these models is that exact