最大后验推断在近似推断中的应用

Application of the maximum posterior inference in approximation inference

文章介绍了几个基本的技巧,用来解决难以处理的推断问题。在深度学习中难以处理的推断问题通常源于结构化图模型中隐变量之间的相互作用。这些相互作用可能是无向模型的直接作用,也可能是有向模型中同一个可见变量的共同祖先之间的explaining away作用。

This article introduces several basic techniques to solve difficult inferential problems. Inferential problems that are difficult to deal with in deep learning are usually derived from the interaction between implicit variables in the structured graph model. These interactions may be the direct effect of the undirected model, or may be the interaction between the common ancestor of the same visible variable in the model.