权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Distributed and Integrated Learning Algorithm for Probabilistic Latent Variable Model

概率潜变量模型的分布式集成学习算法

基本信息

批准号：
24700135
负责人：
SATO Issei
金额：
$ 2.83万
依托单位：
The University of Tokyo
依托单位国家：
日本
项目类别：
Grant-in-Aid for Young Scientists (B)
财政年份：
2012
资助国家：
日本
起止时间：
2012-04-01 至 2014-03-31
项目状态：
已结题

项目摘要

Probabilistic latent variable models have attracted attention in many scientific fields because of their power and flexibility to model real world phenomena.Latent variable reveal the the underlying structure in data. For example, a probabilistic latent variable model for network such as social network enables researchers to analyze latent community in a network. However, learning probabilistic latent variable model is difficult. Typically, learning probabilistic latent variable model is formulated by an optimization problem which has many poor local solutions. We provided an efficient two learning algorithms to find better local solutions. One is based on a collapsed variational Bayes inference, which is a deterministic algorithm. Another is based on a stochastic search with quantum annealing, which is a stochastic algorithm. We found that these algorithms outperformed existing methods in an academic paper analysis analysis and a network data anaysis.

概率隐变量模型因其对真实的世界现象建模的能力和灵活性而受到许多科学领域的关注，隐变量揭示了数据的潜在结构。例如，用于网络（诸如社交网络）的概率潜在变量模型使研究人员能够分析网络中的潜在社区。然而，学习概率潜变量模型是困难的。通常，学习概率潜变量模型是由一个优化问题，其中有许多不良的局部解。我们提供了一个有效的学习算法，以找到更好的局部解决方案。一个是基于一个崩溃的变分贝叶斯推理，这是一个确定性的算法。另一种是基于量子退火的随机搜索，这是一种随机算法。通过对一个学术论文的分析和一个网络数据的分析，我们发现这些算法的性能优于现有的方法。