一种基于最大熵模型的加权归纳迁移学习方法  

A Weighted Algorithm of Inductive Transfer Learning Based on Maximum Entropy Model

作　　者：梅灿华[1] 张玉红[1] 胡学钢[1] 李培培[1] 

机构地区：[1]合肥工业大学计算机科学与技术系,合肥230009

出　　处：《计算机研究与发展》2011年第9期1722-1728,共7页Journal of Computer Research and Development

基　　金：国家"九七三"重点基础研究发展计划基金项目(2009CB326203);国家自然科学基金项目(60975034);安徽省自然科学基金项目(090412044)

摘　　要：传统机器学习和数据挖掘算法主要基于两个假设:训练数据集和测试数据集具有相同的特征空间和数据分布.然而在实际应用中,这两个假设却难以成立,从而导致传统的算法不再适用.迁移学习作为一种新的学习框架能有效地解决该问题.着眼于迁移学习的一个重要分支——归纳迁移学习,提出了一种基于最大熵模型的加权归纳迁移学习算法WTLME.该算法通过将已训练好的原始领域模型参数迁移到目标领域,并对目标领域实例权重进行调整,从而获得了精度较高的目标领域模型.实验结果表明了该算法的有效性.Traditional machine learning and data mining algorithms mainly assume that the training and test data must be in the same feature space and follow the same distribution. However, in real applications, the data distributions change frequently, so those two hypotheses are hence difficult to hold. In such cases, most traditional algorithms are no longer applicable, because they usually require re-collecting and re-labeling large amounts of data, which is very expensive and time consuming. As a new framework of learning, transfer learning could effectively solve this problem by transferring the knowledge learned from one or more source domains to a target domain. This paper focuses on one of the important branches in this field, namely inductive transfer learning. Therefore, a weighted algorithm of inductive transfer learning based on maximum entropy model is proposed. It transfers the parameters of model learned from the source domain to the target domain, and meanwhile adjusts the weights of instances in the target domain to obtain the model with higher accuracy. And thus it could speed up learning process and achieve domain adaptation. The experimental results show the effectiveness of this algorithm.

关 键 词：机器学习 数据挖掘 迁移学习 最大熵 归纳式 ADABOOST 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]