面向张量数据的迁移学习算法

摘要

Most of the existing transfer learning algorithms have been extersively studied based on the vector data .We may transfer a higher order tensor data into a vector-based data as input , and utilize the transfer learning algorithm for vector-based data in higher order tensor data analysis .However , this way will often result in the loss of tensor data '' s space information and the curse of dimensionality problems , so it is necessary to study the transfer learning algorithm based on tensor data .In view of traditional vector-based transfer learning algorithm , this paper presents a transfer learning algorithm based on tensor data .This method makes us to directly use tensor data as an input .In this way we solve the above problems , but also improve the classification ac-curacy .Experiments show that our method can obtain higher classification accuracy compared with others .%目前虽然迁移学习算法得到广泛的研究,但大部分迁移学习算法只是面向向量数据. 面向向量的迁移学习算法首先将高阶的张量数据转换为向量作为输入,再进行处理. 可是在数据类型转换的过程中往往会造成张量数据空间信息的丢失以及维数灾难等问题,因此对张量数据的迁移学习算法的研究显得很有必要. 针对传统的基于向量的迁移算法,本文提出基于张量数据的迁移学习算法,使得可以对张量数据直接作为输入,不仅解决了上述问题,还提高了分类准确率. 实验结果表明,该算法具有较高的分类准确率,有一定的实用价值.