一种新颖的领域自适应概率密度估计器

PDF下载HTML

Title:

A probability density estimator for domain adaptation

作者:

许敏^1,2, 俞林²

1. 江南大学 数字媒体学院, 江苏 无锡 214122;
2. 无锡职业技术学院 物联网技术学院, 江苏 无锡 214121

Author(s):

XU Min^1,2, YU Lin²

1. School of Digital Media, Jiangnan University, Wuxi 214122, China;
2. School of Internet of Things Technology, Wuxi Institute of Technology, Wuxi 214121, China

关键词:

概率密度函数; 无偏置 v-SVR; 中心约束最小包含球; 核心集; 领域自适应

Keywords:

probability density estimation; no bias v-SVR; center-constrained minimum enclosing ball(CC-MEB); core set; domain adaptation

分类号:

TP391.4

DOI:

10.3969/j.issn.1673-4785.201312041

文献标志码:

A

摘要:

传统概率密度估计法建立好密度估计模型后,无法将源域知识传递给相关目标域密度估计模型。提出用无偏置 v-SVR 的回归函数来表示传统概率密度估计法获得密度估计信息,并说明无偏置 v-SVR 等价于中心约束最小包含球及概率密度回归函数可由中心约束最小包含球中心点表示。在上述理论基础上提出中心点知识传递领域自适应概率密度估计法,用于解决因目标域信息不足而无法建立概率密度函数的场景。实验表明,此种领域自适应方法进行领域间知识传递的同时,还能达到源域隐私保护的目的。

Abstract:

This paper proposes that the density information received from the traditional probability density estimation method can be represented by no bias v-SVR regression function. It addresses the problem that after the source domain’s probability density estimation model is established using the traditional probability density estimation method its source domain knowledge can not be transferred to the relevant target domain’s density estimation model. In this paper, no bias v-SVR is equivalent to the center-constrained minimum enclosing ball (CC-MEB) and the probability density regression function is constrained by CC-MEB’s center point is described. On the basis of the above theory, an adaptive probability density evaluation method for transferring knowledge through the center point was put forward to solve the problem that an accurate probability density estimation model can not be established because of the lack of information of the target domain. The experiments showed that this adaptive method can reach the goals of knowledge transfer between domains and privacy protection in the source domain.