Classifications with transferred samples based on RF-spaces
Lin Xiong, Licheng Jiao, Fei Yin · 2014
In supervised learning, it is difficult to gain a good learner when samples with given labels are scarce. Moreover, there are many samples with labels which have a different distribution from the training set in the real environment. In order to improve the learning performance based on scarce training samples, a novel method is proposed that classification with transferred samples in RF-spaces. In the proposed method, transfer learning is introduced to deal with the classification problem with lacking training samples. Based on its mind, the proposed method selects some available samples effectively from the set of samples with different distribution by comparing them with the training samples. Specifically, samples with different distribution are divided into different subsets, and then they are projected on RF-spaces. Finally, the most similar subset is selected and added into the training set to improve the classification performance. Experimental results of UCI and Text datasets illustrate that the proposed method obtains the better classification performance than other methods.