Bayesian网中概率参数学习方法

Bayesian网已经成为AI领域的研究热点,并在现代专家系统、诊断系统及决策支持系统中发挥着至关重要的作用.Bayesian网的研究主要集中在三个方面:知识表示、学习与推理.概率知识是Bayesian网坚实的数学基础,从数据中学习分布参数使得Bayesian网逐步走向现实应用.本文介绍和比较了概率参数学习的各种常用方法,并探求了它们在不同应用背景下的优缺点.基于经典统计学的方法理论成熟,计算简单,但它只利用了实例数据集合所提供的信息,无法加入专家知识,对实例数据的依赖性大;基于Bayesian有机结合了两类信息,对实例数据的依赖性降低,学习结果更加准确.参数学习是Bayesian网学习的基础,是Bayesian网结构学习必不可少的部分.

Learning with a Bayesian Networks a Set of Conditional Probility Tables

Bayesian network is becoming more remarkable in AI research fields,which plays important role in modern expert system,diagnoses system and decision system.Bayesian network works on three points as below:knowledge representation,learning and inference.Prababilistic methods are its mathematical fundamental which helps learning distribution from data and leads Bayesian theory to real application.This paper introduces various common methods in probability data learning and make comparison among them under various application background.The methods based on classical statistics have a matured theory and a set of simple and direct calculation.But they reply heavily on sample data,which apply only those information from sample data with expert knowledge left aside.Bayesian Network combines information of expert knowledge and sample data together.It can give more accurate learning result and rely less on sample data.Parameter learning is main part of learning Bayesian Network models,and it's the basis of Bayesian Network learning.