消息传递方法及其在信息融合中的应用

作为针对复杂问题的确定性推断方法,近年来消息传递方法及其在信息融合领域中的应用备受关注.消息传递为大规模问题提供了基于贝叶斯的、统一的、可扩展的、效率高的推断框架,消息传递方法在概率图模型的节点之间传递消息.首先,对概率图模型进行简要介绍,总结消息传递基础理论方法和近似拓展方法的基本原理、特点以及研究进展;然后,针对信息融合领域中的3类重难点问题:状态估计与平滑、目标跟踪和多源异构数据融合,分别综述消息传递在信息融合中的最新研究进展,并总结不同消息传递方法的适用场景;最后,讨论消息传递方法在信息融合领域未来可能的研究方向.

Message passing methods and their applications in information fusion

As a deterministic inference method for complicated problems, message passing methods and their applications in information fusion have drawn much attention in recent years. Message passing methods provide a Bayes-based, unified, scalable and efficient inference framework for large scale problems. Message passing methods pass messages between nodes of probabilistic graphical models. At first, probabilistic graphical models are briefly introduced. Then, the basic principles and a brief review of common message passing methods are given. Then, the recent applications of message passing methods in information fusion are presented from three aspects: state estimation and smoothing, target tracking and multi­source heterogeneous data fusion. Meanwhile, the appropriate scenarios of different message passing methods are summarized. Finally, the possible directions of future research of message passing methods in information fusion are discussed.