基于迁移共享空间的分类新算法

发布时间：2018-05-02 20:34

本文选题：共享空间 + 迁移学习　；参考：《计算机研究与发展》2016年03期

【摘要】：为解决来自不同但相关领域的大量无标签数据和少量带标签数据的分类问题,首先构造一个联系源域到目标域的共享特征空间,并将该空间引入经典的支持向量机算法使其获得迁移能力,最终得到一种新的基于支持向量机的迁移共享空间的分类新算法,即迁移共享空间支持向量机.具体地,该方法以迁移学习理论为基础,结合分类器最大间隔原理,通过最大化无标签数据和带标签数据的联合概率分布来构建无标签数据和带标签数据的共享空间;为充分考虑少量带标签数据之数据分布,在其原始特征空间和共享空间组成的扩展空间中训练分类模型.相关实验结果验证了该迁移学习分类器的有效性.
[Abstract]:In order to solve the classification problem of a large number of untagged data and a small amount of tagged data from different but related domains, a shared feature space from the source domain to the target domain is constructed. The space is introduced into the classical support vector machine (SVM) algorithm to obtain the migration ability. Finally, a new classification algorithm based on support vector machine (SVM) is proposed, that is, the migration shared space support vector machine (SVM). Specifically, based on the transfer learning theory and the principle of classifier maximum interval, the shared space of untagged data and tagged data is constructed by maximizing the joint probability distribution of untagged data and tagged data. In order to fully consider the data distribution with a small amount of tagged data, the classification model is trained in the extended space composed of the original feature space and the shared space. The experimental results demonstrate the effectiveness of the proposed transfer learning classifier.
【作者单位】：江南大学数字媒体学院;齐鲁工业大学信息学院;
【基金】：国家自然科学基金项目(61170122,61202311) 山东省高等学校科技计划基金项目(J14LN05)~~
【分类号】：TP18

【相似文献】