Variational Bayesian Interpretaion of the Kalman Filter

摘要：

Kalman filter is regarded as the optimal solution to many estimation tasks for linear/Gaussian systems. There are several intuitive explanations of kalman filter in the literature, such as a mean squared error minimiser, likelihood maximumer. From the point of view of information theory, the estimation of posterior distribution in Bayesian estimation was considered as identical with optimization problem of information. This leads to the variational Bayesian theory which combined Bayesian estimation with calculus of variants which deals with extremizing functionals. Apparent information is introduced as a score function in this paper, and variational Bayesian estimation for linear/Gaussian discrete system is studied. The corresponding deduction obtained result which is equal to Kalman filter equations, thus got a different interpretation of Kalman filter from the perspective of variational Bayes. In addition, the result of our deduction simplified the equations of Kalman filter.

关键词： Kalman filtering variational Bayes apparent information

作者: YAO Zhi-ying LIU Dong

作者单位: Xian High-tech Institute 303 Staff Room Xian, China

会议类型: 国际会议

会议名称: 2010 International Conference on Measurement and Control Engineering(2010年IEEE测量与控制工程国际会议 ICMCE2010)

会议地点: 成都

会议语种:英文

页码: 107-110

在线出版日期: 2010-11-16（万方平台首次上网日期，不代表论文的发表时间）