CAJ | 학술논문

基于密度差(Difference Of Density, DOD)思想，L2核分类器算法具有良好的分类性能及稀疏性，然而其训练域与测试域独立同分布的假设限制了其应用范围。针对此不足，该文提出一种新的面向迁移学习的L2核分类器(Transfer Learning-L2 Kernel Classification, TL-L2KC)，该方法既保持了L2核分类器算法良好的分类性能，又能处理数据集缓慢变化及训练集在特定约束条件下获得导致训练集和未来测试集分布不一致的问题。基于人造数据集和UCI真实数据集的实验表明，该文提出的TL-L2KC算法较之于经典的迁移学习分类方法，具有相当的、甚至更好的性能。
기우밀도차(Difference Of Density, DOD)사상，L2핵분류기산법구유량호적분류성능급희소성，연이기훈련역여측시역독립동분포적가설한제료기응용범위。침대차불족，해문제출일충신적면향천이학습적L2핵분류기(Transfer Learning-L2 Kernel Classification, TL-L2KC)，해방법기보지료L2핵분류기산법량호적분류성능，우능처리수거집완만변화급훈련집재특정약속조건하획득도치훈련집화미래측시집분포불일치적문제。기우인조수거집화UCI진실수거집적실험표명，해문제출적TL-L2KC산법교지우경전적천이학습분류방법，구유상당적、심지경호적성능。
Based on the concept of Difference Of Density (DOD), L2 Kernel Classifier(L2KC) exhibits its good performance. However, the assumption that the training domain and testing domain are independent and identically distributed severely constrains its usefulness. In order to overcome this shortcoming, a novel classifier named Transfer Learnging-L2 Kernel Classification (TL-L2KC) is proposed for the changing environment. The proposed classifier can not only inherit the advantage of L2KC, but also deal with the problem that the distribution inconsistency between the training and testing sets which is caused by the slow change of the datasets or the training set obtained with specific constraints. As demonstrated by extensive experiments in simulation datasets and UCI benchmark datasets, the proposed classifier TL-L2KC shows the performance which is comparable to or better than that of the classical algorithms on the transfer learning classification problems.