摘要
本文针对狄利克雷混合模型提出了有效的随机变分推理算法。首先,构建抽样数据的变分目标函数的下界;其次,利用随机优化和自然梯度下降算法推导出变分后验分布的解析解表达式;最后,将其应用于图像分类问题,实验验证了该算法的有效性。
This paper proposes an efficient stochastic variational inference algorithm for the Dirichlet mixture model.Firstly,the lower bound of the variational objective function of the sampled data is constructed;Secondly,the analytical solution expression of the variational posterior distribution is solved by using stochastic optimization and natural gradient descent algorithm;Finally,the proposed algorithm is applied for image categorization,which verifies the effectiveness and feasibility of the algorithm.
作者
曹会蕊
关文博
杨帆
CAO Hui-rui;GUAN Wen-bo;YANG Fan(School of Information Science and Technology,North China University of Technology,Beijing 100144)
出处
《数字技术与应用》
2021年第9期109-111,共3页
Digital Technology & Application
关键词
图像分类
混合模型
狄利克雷分布
随机变分推理
Image classification
Mixture model
Dirichlet distribution
Stochastic variational inference