Distributed EM algorithm for Gaussian mixtures in sensor networks

被引:120
作者
Gu, Dongbing [1 ]
机构
[1] Univ Essex, Dept Comp & Elect Syst, Colchester CO4 3SQ, Essex, England
来源
IEEE TRANSACTIONS ON NEURAL NETWORKS | 2008年 / 19卷 / 07期
关键词
consensus filter; distributed estimation; distributed expectation-maximization (EM) algorithm; sensor networks;
D O I
10.1109/TNN.2008.915110
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
This paper presents a distributed expectation-maximization (EM) algorithm over sensor networks. In the E-step of this algorithm, each sensor node independently calculates local sufficient statistics by using local observations. A consensus filter is used to diffuse local sufficient statistics to neighbors and estimate global sufficient statistics in each node. By using this consensus filter, each node can gradually diffuse its local information over the entire network and asymptotically the estimate of global sufficient statistics is obtained. In the M-step of this algorithm, each sensor node uses the estimated global sufficient statistics to update model parameters of the Gaussian mixtures, which can maximize the log-likelihood in the same way as in the standard EM algorithm. Because the consensus filter only requires that each node communicate with its neighbors, the distributed EM algorithm is scalable and robust. It is also shown that the distributed EM algorithm is a stochastic approximation to the standard EM algorithm. Thus, it converges to a local maximum of the log-likelihood. Several simulations of sensor networks are given to verify the proposed algorithm.
引用
收藏
页码:1154 / 1166
页数:13
相关论文
共 30 条
[1]   A survey on sensor networks [J].
Akyildiz, IF ;
Su, WL ;
Sankarasubramaniam, Y ;
Cayirci, E .
IEEE COMMUNICATIONS MAGAZINE, 2002, 40 (08) :102-114
[2]   A tutorial on particle filters for online nonlinear/non-Gaussian Bayesian tracking [J].
Arulampalam, MS ;
Maskell, S ;
Gordon, N ;
Clapp, T .
IEEE TRANSACTIONS ON SIGNAL PROCESSING, 2002, 50 (02) :174-188
[3]   Fastest mixing Markov chain on a graph [J].
Boyd, S ;
Diaconis, P ;
Xiao, L .
SIAM REVIEW, 2004, 46 (04) :667-689
[4]   EM procedures using mean field-like approximations for Markov model-based image segmentation [J].
Celeux, G ;
Forbes, F ;
Peyrard, N .
PATTERN RECOGNITION, 2003, 36 (01) :131-144
[5]   Unsupervised learning of Gaussian mixtures based on variational component splitting [J].
Constantinopoulos, Constantinos ;
Likas, Aristidis .
IEEE TRANSACTIONS ON NEURAL NETWORKS, 2007, 18 (03) :745-755
[6]   MAXIMUM LIKELIHOOD FROM INCOMPLETE DATA VIA EM ALGORITHM [J].
DEMPSTER, AP ;
LAIRD, NM ;
RUBIN, DB .
JOURNAL OF THE ROYAL STATISTICAL SOCIETY SERIES B-METHODOLOGICAL, 1977, 39 (01) :1-38
[7]   A spatially constrained generative model and an EM algorithm for image segmentation [J].
Diplaros, Aristeidis ;
Vlassis, Nikos ;
Gevers, Theo .
IEEE TRANSACTIONS ON NEURAL NETWORKS, 2007, 18 (03) :798-808
[8]  
Doucet A., 2001, SEQUENTIAL MONTE CAR, V1, DOI [10.1007/978-1-4757-3437-9, DOI 10.1007/978-1-4757-3437-9]
[9]  
Estrin D., 1999, P MOBICOM, DOI DOI 10.1145/313451.313556
[10]   Unsupervised learning of finite mixture models [J].
Figueiredo, MAT ;
Jain, AK .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2002, 24 (03) :381-396