基于Logistic回归分析的直推式迁移学习  

Logistic regression for transductive transfer learning

作　　者：胡学钢[1] 方玉成[1] 张玉红[1] 

机构地区：[1]合肥工业大学计算机与信息学院,安徽合肥230009

出　　处：《合肥工业大学学报（自然科学版）》2010年第12期1797-1801,1810,共6页Journal of Hefei University of Technology：Natural Science

基　　金：国家自然科学基金资助项目(60975034);安徽省自然科学基金资助项目(090412044);合肥工业大学科学研究发展基金资助项目(2010HGXJ0013)

摘　　要：传统的机器学习方法基于一个基本的假设:训练数据和测试数据遵循相同的分布。然而,在许多现实的应用中,这种假设并不能够被保证。在这种情况下,传统的机器学习方法因没有意识到分布的改变而可能失败。近年来,迁移学习技术被专门用来解决这一缺陷。文章提出了一种叫做TTLR的方法,将原始领域中的训练数据有效地迁移到目标领域中,该方法首先对Logistic回归分析模型进行扩展,然后利用不同领域概率分布之间的差异性,调节训练数据中每个实例的权重,从而使得训练得到的分类器更加适应于目标领域;在所选取的数据集上得到的实验结果表明,与传统的监督式学习方法相比,所提出的方法有很大的优势。Traditional machine learning makes a basic assumption that the training data and test data should be under the same distribution.However,in many practical applications,this assumption does not hold.In such situations,traditional machine learning methods may not work because of its unawareness of the shift of distribution.In recent years,transfer learning techniques are purposed to address this shortcoming.A novel method named transductive transfer learning regression（TTLR） is proposed in this paper,which allows training data to be effectively transferred from the source domain to the target domain.First,the proposed method extends the logistic regression model.Then,it exploits the difference of the distributions in different domains to adjust the weights of the instances in the training data so that the source labeled data is more adaptive to the target domain.Experimental results on the selected datasets indicate that the method outperforms the traditional supervised learning one.

关 键 词：机器学习 LOGISTIC回归分析 直推式迁移学习 分类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]