Learning Probabilistic Networks from Databases

从数据库学习概率网络

• 批准号: 9111590
• 负责人: Gregory Cooper
• 金额: $ 10.81万
• 依托单位: University of Pittsburgh
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1991
• 资助国家: 美国
• 起止时间: 1991-09-01 至 1994-02-28
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9111590&HistoricalAwards=false
• 关键词: Learning; Probabilistic; Networks; Databases

This research investigates Bayesian methods for constructing probabilistic networks from databases. Its main focus is on constructing Bayesian belief networks. Primary goals are to (1) develop methods for calculating the probability of a Bayesian belief-network structure given a database of cases, (2) identify the most probable belief-network structure given a database of cases, and (3) perform probabilistic inference by taking a weighted average over the inferences of multiple belief networks. Potential applications include computer-assisted hypothesis testing, automated scientific discovery, and automated construction of probabilistic expert systems. Methods will be explored for intergrateing prior knowledge with data, handling missing data, and discovering hidden (latent) variables. Of particular concern is the development of computationally efficient algorithms. The methods developed will be empirically evaluated using databases from several domains. //

这项研究调查了用于从数据库构建概率网络的贝叶斯方法。 它的主要重点是构建贝叶斯信仰网络。 主要目标是（1）开发用于计算案例数据库的贝叶斯信仰网络结构概率的方法，（2）在给定情况数据库的情况下确定最可能的信仰网络结构，以及（3）执行概率推理通过对多重信仰网络的推论进行加权平均值。 潜在的应用包括计算机辅助的假设测试，自动化科学发现以及概率专家系统的自动构造。 将探讨方法，以将先验知识与数据相互汇总，处理丢失的数据并发现隐藏（潜在）变量。 特别关注的是计算有效算法的发展。 开发的方法将使用来自多个域的数据库进行经验评估。 //