Abstract: A Gaussian Fields(GF) on nearest neighbor graph is defined by using a non-parametric technique. On the basis of it, a MAP criterion which can automatically set model parameter and numbers of nearest-neighbor k is proposed and entropy maximization query selection method for active learning by using supervised and unsupervised information is specified. Experimental results demonstrate effectiveness of GF compared with semi-active learning method.

Key words: Gaussian fields(GF), semi-supervised regression, active learning, entropy, Cholesky decomposition

摘要： 介绍一种定义近邻图上的高斯域(GF)及用于降维和分类的GF的相关知识，提出一种用于半监督回归的高斯域，能自动设置模型参数和近邻数，利用监督和无监督数据进行熵值查询选择从而进行主动学习。实验将其与半监督学习法进行比较并验证了GF的有效性。

关键词: 高斯域, 半监督回归, 主动学习, 熵, Cholesky分解

CLC Number: 

• TP391