首页 | 本学科首页   官方微博 | 高级检索  
     

相容的贝叶斯学习及其后验分布的渐近正态性
引用本文:胡振宇,林士敏,陆玉昌. 相容的贝叶斯学习及其后验分布的渐近正态性[J]. 广西师范大学学报(自然科学版), 2004, 22(1): 43-46
作者姓名:胡振宇  林士敏  陆玉昌
作者单位:中国科学院,软件研究所,北京,100080;广西师范大学,数学与计算机科学学院,广西,桂林,541004;清华大学,计算机科学与技术系,北京,100084
基金项目:清华大学智能技术与系统国家重点实验室开放课题资助项目(99002)
摘    要:从计算学习理论的角度研究贝叶斯学习的相容性和后验分布的渐近正态性,给出了贝叶斯学习的正则条件,证明了在这些条件下贝叶斯学习不仅是相容的,而且后验分布是渐近正态的.由于正态分布计算相对简单,该结果为指派恰当有效的先验分布、寻找简化贝叶斯学习计算的方法提供了理论依据,给出的正则条件比Heyde与Johnstone的5个条件更为简化,便于应用.

关 键 词:机器学习  贝叶斯学习  相容性  渐近正态性  数据采掘
文章编号:1001-6600(2004)01-0043-04

THE ASYMPTOTIC NORMALITY OF POSTERIOR IN BAYESIAN LEARNING
HU Zhen-yu,LIN Shi-min,LU Yu-chang. THE ASYMPTOTIC NORMALITY OF POSTERIOR IN BAYESIAN LEARNING[J]. Journal of Guangxi Normal University(Natural Science Edition), 2004, 22(1): 43-46
Authors:HU Zhen-yu  LIN Shi-min  LU Yu-chang
Affiliation:HU Zhen-yu~1,LIN Shi-min~2,LU Yu-chang~3
Abstract:This paper studies the consistency and asymptotic normality of posterior in Bayesian learning.It presents a set of regular conditions for Bayesian learning,and proves that under these conditions Bayesian learning has not only consistency but also has normal distribution of posterior asymptotically.Because the computing of normal distribution is relatively simple,the results in this paper provide a theoretic basis for assessing resultful prior and methods to reduce the computing in Bayesian learning.The regular conditions presented in this paper are simpler than the 5 conditions given by Heyde and Johnstone,and more suitable for application.
Keywords:machine learning  Bayesian learning  posterior distribution  asymptotic normality  data mining
本文献已被 CNKI 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号