二值probit回归模型的坍缩变分贝叶斯推断算法  

Collapsed variational Bayesian inference algorithm for binary probit regression model

作　　者：卿湘运[1] 王行愚[1] 牛玉刚[1] 

机构地区：[1]华东理工大学信息科学与工程学院,上海200237

出　　处：《控制与决策》2008年第5期589-592,共4页Control and Decision

基　　金：国家973计划项目(2002CB312203);国家自然科学基金项目(60674089);上海市重点学科研究项目(B504)

摘　　要：给出了二值probit回归模型的坍缩变分贝叶斯推断算法.此算法比变分贝叶斯推断算法能更逼近对数边缘似然,得到更精确的模型参数后验期望值.如果两个算法得到的分类错误一致,则该算法的迭代次数较变分法明显减少.仿真实验结果验证了所提出算法的有效性.A collapsed variational Bayesian inference algorithm for binary probit regression model is proposed. The algorithm is a better approximation of the alogrithm of the marginal likelihood than standard variational Bayesian inference algorithm. It also can achieve more accurate posterior expectation of model parameters than the latter. If the classification error rates of two algorithms are dose, the iteration number of the proposed algorithm is siginifieantly decreased. Simulation experiments show the effectiveness of the algorithm.

关 键 词：二值probit模型 变分贝叶斯推断 坍缩变分贝叶斯推断 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]