安全迁移支持向量机-计算机科学.PDF

第 44 卷第11A 期 Vo l. 44 No. 11A 计算机科学 2017 年 11 月 COMPUTER SCIENCE Nov.2017 安全迁移支持向量机 周国华巢海鲸申燕萍 (常州轻工职业技术学院信息工程系 常州 213164) 摘 要 迁移学习方法是一种新的机器学习框架，它将源领域数据通过学习迁移到相似的目标领域中，减弱了对已标 记数据的依赖。但迁移学习方法中一个重大问题是使用目标领域数据与源领域数据得到的分类器很可能比仅利用目 标领域数据得到的分类器的效果更差，从而造成一种负迁移现象。针对此问题，提出一种基于目标领域已标记数据 知识的安全控制机制，并通过结合近年出现的一种迁移学习分类器CTL-SVM) 提出了一种安全迁移支持向量机 (SATL-SVM) ，从理论上解决了 TL-SVM 的负迁移问题，在人工数据集和真实数据集上的实验结采表明了所提方法 的有效性。 关键词 迁移学习，支持向量机，分类 中固法分类号 TP39 1. 4 文献标识码 A Safety-aware Transfer Learning Support Vector Machine ZHOU Guo-hua CHAO Hai-jing SHEN Yan-ping (Department of Information Engineering ,Changzhou Institute of Light Industry Technology ,Changzhou 213164 ,China) Abstract Transfer learning method is a new machine learning framework ,which aims at realizing an effective learning for target domain by efficiently transferring the existing knowledge from source domain to target domain. It would be nice to reduce the need and effort to recollect the training data. However ,the classifier trained by the data of source do main and target domain often put in a worse performance than the classifier only trained by the data of target domain , which leads to the appearance of negative transfer. To address this problem , a safety-control mechanism with the knowledge of the labeled data of target domain was proposed for safe transfer leaning. Furthermore ,in implementation , based on a recent transfer learning method (TL-SVM) ,a safety-aware transfer learning support vector machine(STL SVM) was proposed ,which avoids the appearance of negative transfer of TL-SVM th