CAJ | 학술논문

为了寻找一种更好的目标分类器，提出了基于子空间插值的领域适应学习方法。从源域出发，依次训练中间领域字典，缩小样本分布差，形成一条平滑的虚拟路径，逐步适应目标数据，达到提高识别性能的目的。在CM U-PIE上进行了实验，结果表明此方法较其他领域适应学习方法有更好的分类性能。
위료심조일충경호적목표분류기，제출료기우자공간삽치적영역괄응학습방법。종원역출발，의차훈련중간영역자전，축소양본분포차，형성일조평활적허의로경，축보괄응목표수거，체도제고식별성능적목적。재CM U-PIE상진행료실험，결과표명차방법교기타영역괄응학습방법유경호적분류성능。
In order to find a better target classifier , this paper proposed the spatial interpolation of domain adaptation method . Starting from the source domain , followed by training intermediate dictionary in the field to further reduce the sampling distribution difference , forming a smooth virtual path , gradually adapting to the target data , achieve the goal of a better performance . And conducted experiment results on the CM U-PIE show that the performance of this method is better than the other areas of adaptive learning method .