《计算机应用研究》|Application Research of Computers

层次K-均值聚类结合改进ITML的迁移度量学习方法

Transfer metric learning method based on hierarchical K-means clustering and improved ITML

免费全文下载 (已被下载 次)  
获取PDF全文
作者 蒋林利,吴建生
机构 1.广西科技师范学院 数学与计算机科学学院,广西 来宾 546199;2.武汉大学 软件工程国家重点实验室,武汉 430072;3.武汉理工大学 信息工程学院,武汉 430070
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2017)12-3552-04
DOI 10.3969/j.issn.1001-3695.2017.12.007
摘要 目前的迁移学习方法多针对单一迁移类型,使用低级特征空间,并且源集比目标集复杂耗力。针对这些问题,综合考虑特征表示迁移、参数迁移和实例迁移,提出迁移度量学习的通用框架。首先,基于属性相似性空间和类别相似性空间,利用层次K-均值聚类获取相似性;然后,利用信任评估框架和去相关归一化转换方法消除源集中的相关关系来抑制负迁移作用;最后,改进信息理论度量学习方法(ITML)进行相似性度量学习。对三种不同复杂度数据集进行实验,结果表明,提出方法的迁移学习性能较传统方法明显提高,且对负迁移影响具有更好的鲁棒性;提出的方法可应用于源集比目标集简单的情况,评估结果表明,即使源集知识有限,也可以得到较好的迁移学习效果。
关键词 迁移度量学习;层次K-均值聚类;相似性空间;信任评估框架;去相关归一化空间;信息理论度量学习
基金项目 国家自然科学基金资助项目(61202143)
广西自然科学基金资助项目(2014GXNSFAA118027)
本文URL http://www.arocmag.com/article/01-2017-12-007.html
英文标题 Transfer metric learning method based on hierarchical K-means clustering and improved ITML
作者英文名 Jiang Linli, Wu Jiansheng
机构英文名 1.SchoolofMathematics&ComputerScience,GuangxiScience&TechnologyNormalUniversity,LaibinGuangxi546199,China;2.StateKeyLaboratoryofSoftwareEngineering,WuhanUniversity,Wuhan430072,China;3.SchoolofInformationEngineering,WuhanUniversityofTechnology,Wuhan430070,China
英文摘要 Now most of transfer learning methods suffer from the problems that transfer types are separately analyzed, low level feature space are used, and the source data set is more diverse and complex than the target set. For these problems, this paper proposed a novel general transfer metric learning framework with comprehensive consideration of feature representation transfer, parameter transfer and instance transfer. Initially, it used hierarchical K-means clustering to get the similarity based on the semantic similarity space and category similarity space. Then, it utilized the trust evaluation framework and de-correlated normalized space to eliminate the correlation learned in the source domain, and restrained the negative transfer. Finally, it modified the information theoretic metric learning to precede similarity metric learning. The experiment results show that the transfer learning performance of the proposed method has improved greatly with more robust to negative transfer effect comparing with the traditional methods in three data sets with different complexity. Furthermore, the proposed method could be applied in the situation that the source data set was simpler than the target set. The results reveal that even when the knowledge source is limited, transfer learning can still be beneficial.
英文关键词 transfer metric learning; hierarchical K-means clustering; similarity space; trust evaluation framework; de-correlated normalized space; information theoretic metric learning(ITML)
参考文献 查看稿件参考文献
  [1] 庄福振, 罗平, 何清, 等. 迁移学习研究进展[J] . 软件学报, 2015, 26(1):26-39.
[2] 赵新杰, 刘渊, 孙剑. 基于迁移学习和D-S理论的网络异常检测[J] . 计算机应用研究, 2016, 33(4):1137-1140.
[3] Gerkmann T. Bayesian estimation of clean speech spectral coefficients given a priori knowledge of the phase[J] . IEEE Trans on Signal Processing, 2014, 62(16):4199-4208.
[4] 宋鹏, 金赟, 查诚, 等. 基于稀疏特征迁移的语音情感识别[J] . 数据采集与处理, 2016, 31(2):325-330.
[5] Bottillo S, Vollaro A, Galli G, et al. Fluid dynamic and heat transfer parameters in an urban canyon[J] . Solar Energy, 2014, 11(12):1-10.
[6] 张倩, 李明, 王雪松, 等. 一种面向多源领域的实例迁移学习[J] . 自动化学报, 2014, 40(6):1176-1183.
[7] Kotzias D, Denil M, Blunsom P, et al. Deep multi-instance transfer learning[J] . Computer Science, 2014, 99(1):550-561. [8] 于重重, 田蕊, 谭励, 等. 非平衡样本分类的集成迁移学习算法[J] . 电子学报, 2012, 40(7):1358-1363.
[9] 张博, 史忠植, 赵晓非, 等. 一种基于跨领域典型相关性分析的迁移学习方法[J] . 计算机学报, 2015, 38(7):1326-1336.
[10] Han Yahong, Wu Fei, Zhuang Yueting, et al. Multi-label transfer learning with sparse representation[J] . IEEE Trans on Circuits and Systems for Video Technology, 2010, 20(8):1110-1121.
[11] Felzenszwalb P, Girshick R, Mcallester D, et al. Object detection with discriminatively trained part-based models[J] . IEEE Trans on Software Engineering, 2014, 32(9):1627-1645.
[12] 危辉, 栾尚敏. 基于连通结构与动力学过程的知觉记忆层模型[J] . 软件学报, 2004, 15(11):1616-1628.
[13] Alhalah Z, Rybok L, Stiefelhagen R. What to transfer? High-level semantics in transfer metric learning for action similarity[C] //Proc of International Conference on Pattern Recognition. 2014:2775-2780.
[14] Liu Yang, Jing Liping, Yu Jian, et al. Learning transferred weights from co-occurrence data for heterogeneous transfer learning[J] . IEEE Trans on Neural Network and Learning Systems, 2015, 27(11):1-10.
[15] 丁艳会, 郝俊寿, 李春明. 基于社团主题的领域相关推荐算法[J] . 湘潭大学自科学报, 2015, 37(4):92-97.
[16] Xiong Feiyu, Kam M, Hrebien L, et al. Kernelized information theoretic metric learning for cancer diagnosis using high-dimensional molecular profiling data[J] . ACM Trans on Knowledge Discovery from Data, 2016, 10(4):1-23.
[17] 周文刚, 赵宇, 朱海. 基于混合高斯模型和空间模糊度的支持向量机算法研究[J] . 计算机应用研究, 2015, 32(5):1319-1321.
收稿日期 2016/12/28
修回日期 2017/2/27
页码 3552-3555,3572
中图分类号 TP182
文献标志码 A