Learning Bayesian Networks that Contain Both Discrete and Continuous Variables

库珀，格雷戈里大学匹兹堡$69,951 - 13。在基于知识的系统中，概率信念网络越来越多地用于对各种领域的信念进行建模。在医学知识和信仰领域进行了一些最广泛的工作，本项目继续了该领域以前的工作。一直困扰概率网络知识库以及其他知识库的问题之一是知识获取。这项工作的目标是利用医学和其他类型的数据库和贝叶斯学习方法来开发概率网络。这里使用的特殊类型的概率网络结构被称为贝叶斯信念网络。所研究的方法分为两个阶段。第一阶段是一种相对通用的技术，用于学习模型变量之间的定性、结构性关系；第二阶段使用特定于分布的技术来量化模型变量之间的关系。在各种模拟和现实世界的数据库上进行测试的假设是，这种方法将产生更具预测性的贝叶斯信念网络模型。

IRI-9509792 Cooper, Gregory University of Pittsburgh $69,951 - 13 mos. Learning Bayesian Networks that Contain both Discrete and Continuous Variables Probabilistic belief networks are increasingly being used to model belief in a great variety of domains in knowledge-based systems. Some of the most extensive work has been done in the area of medical knowledge and beliefs, and this project continues previous work in that domain. One of the problems that has continued to bedevil probabilistic network knowledge bases, as well as other knowledge bases, is knowledge acquisition. The goal in this work is to enable the use of medical and other types of databases and Bayesian learning methods to develop probabilistic networks. The particular type of probabilistic network structure being used here is called a Bayesian belief network. The method being investigated uses two stages. The first stage is a relatively generic technique for learning the qualitative, structural relationships among the model variables; and the second stage uses distribution- specific techniques for quantifying the relationships among the model variables. The hypothesis, to be tested on a variety of simulated and real-world databases, is that this method will produce more predictively accurate Bayesian belief network models.