DocumentCode :
477749
Title :
Coupling and Accuracy of Gaussian Mean Fields
Author :
Chen, Yarui ; Liao, Shizhong
Author_Institution :
Sch. of Comput. Sci. & Technol., Tianjin Univ., Tianjin
Volume :
2
fYear :
2008
fDate :
18-20 Oct. 2008
Firstpage :
24
Lastpage :
28
Abstract :
Gaussian mean field is an elementary variational inference method based on disjoint variable clusters, and its variational accuracy is a critical property for analyzing and applying the variational inference method. In this paper, we explore the relationship between variational accuracy and variable cluster dependence for Gaussian mean field. First, we propose two concepts, named as model coupling and quasi-coupling, to measure the dependence among variable clusters. Then, we prove a theorem regarding the quantitative relationship between variational accuracy and model coupling. We also analyze the qualitative relationships between model coupling and quasi-coupling, and between variational accuracy and quasi-coupling. Finally, we design experiments to demonstrate the theoretical results.
Keywords :
Gaussian processes; Markov processes; inference mechanisms; variational techniques; Gaussian mean fields; disjoint variable clusters; elementary variational inference method; model coupling; quasicoupling; Computer science; Exponential distribution; Fuzzy systems; Graphical models; Markov random fields; Random variables; Statistical analysis; Statistical distributions; Symmetric matrices; Weight measurement;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Fuzzy Systems and Knowledge Discovery, 2008. FSKD '08. Fifth International Conference on
Conference_Location :
Shandong
Print_ISBN :
978-0-7695-3305-6
Type :
conf
DOI :
10.1109/FSKD.2008.572
Filename :
4666073
Link To Document :
بازگشت