arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1708.06901cs.ITmath.IT
收藏

参数未知时贝叶斯最快变点检测的多图检测方法

Multi-Chart Detection Procedure for Bayesian Quickest Change-Point Detection with Unknown Post-Change Parameters

Jun Geng, Erhan Bayraktar, Lifeng Lai

摘要:

在贝叶斯框架下研究了随机过程突变的快速检测问题。与经典的贝叶斯最快变点检测问题不同,本文考虑了变化后分布不确定的情况。具体地说,观察者只知道变化后的分布属于参数分布族,但他不知道变化后参数的真实值。在这种情况下,我们提出了两个多图检测过程,分别称为M-SR过程和修改的M-SR过程,并证明了这两个过程是渐近最优的,当后改变的参数属于一个有限集,并渐近$ε-$最优的后改变的参数属于一个有限的措施紧集。这两种算法都可以有效地计算,因为它们的检测统计数据可以递归地更新。然后,我们扩展的研究,考虑多源监测问题与未知的变化后的参数。当这些监测源是相互独立的,我们提出了一个基于窗口的修改M-SR检测过程,并表明,所提出的检测方法是一阶渐近最优的后变化的参数属于有限集。我们表明,所提出的算法的计算和空间复杂度只线性增加相对于源的数量。

英文摘要:

In this paper, the problem of quickly detecting an abrupt change on a stochastic process under Bayesian framework is considered. Different from the classic Bayesian quickest change-point detection problem, this paper considers the case where there is uncertainty about the post-change distribution. Specifically, the observer only knows that the post-change distribution belongs to a parametric distribution family but he does not know the true value of the post-change parameter. In this scenario, we propose two multi-chart detection procedures, termed as M-SR procedure and modified M-SR procedure respectively, and show that these two procedures are asymptotically optimal when the post-change parameter belongs to a finite set and are asymptotically $ε-$optimal when the post-change parameter belongs to a compact set with finite measure. Both algorithms can be calculated efficiently as their detection statistics can be updated recursively. We then extend the study to consider the multi-source monitoring problem with unknown post-change parameters. When those monitored sources are mutually independent, we propose a window-based modified M-SR detection procedure and show that the proposed detection method is first-order asymptotically optimal when post-change parameters belong to finite sets. We show that both computation and space complexities of the proposed algorithm increase only linearly with respect to the number of sources.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑