Logistic混合模型的变分贝叶斯推断
Variational Bayesian Inference for Logistic mixture model
龚斌 1赵凝 1郑静1
作者信息
- 1. 杭州电子科技大学经济学院,浙江杭州 310018
- 折叠
摘要
传统的变分贝叶斯推断(Variational Bayesian Inference)面对无共轭先验分布的混合模型无法进行有效的参数估计,Logistic混合模型即为其中的一种,对此模型,提出一种新的变分贝叶斯算法.该算法通过将Logistic函数进行近似表示来解决无共轭分布的问题,进而实现该类混合模型的参数估计.首先给出变分贝叶斯推断的完整推导,并在此基础上提出变分贝叶斯推断算法,解决该混合模型的参数估计问题;最后,在数据集上进行实证分析,结果表明,该算法能在精确估计的同时实现快速收敛.
Abstract
The traditional variational Bayesian inference method fails to perform effective parameter estimation when inferring non-conjugate mixed models.Therefore,a new variational Bayesian algorithm is proposed.This algorithm solves the problem of not forming conjugate distribution by approximating the Logistic function,and then realizes the parameter estimation of this kind of mixed model.Firstly,the component is fixed and the prior setting is given.Then,the complete derivation of the variational Bayesian inference is given,and on this basis,the Variational Bayesian Inference algorithm is proposed to solve the problem of parameter estimation of the mixed model.Finally,an empirical analysis is carried out on the dataset.The corresponding and the results show that the algorithm can achieve fast convergence for its accurate estimation.
关键词
变分推断/混合模型/参数估计Key words
Variational inference/Mixture model/Parameter estimation引用本文复制引用
出版年
2024
NETL
NSTL
万方数据