A Transfer Learning Algorithm Based on Support Vector Machine

In many scenarios of classification, test and training data must come from same feature space and have same distribution. However, this assumption may not be satisfactory in many practical applications. In recent years, the emergence of transfer learning has provided a new technology to solve this problem. Aiming at problems that the existing transfer learning algorithms only utilize the data in source domain to assist the learning tasks of target domain without considering the effect of target data to source domain, the selectivity bias of knowledge and scarce labeled samples, this paper proposed a collaboration mutual transfer learning algorithm based on support vector machine -CMATLSVM.CMATLSVM realizes multiple mutual learnings between source and target domains through the collaborative mutual assistance. In collaborative mutual assistance, in order to further improve the classification effect of target domain task, a similarity measure is introduced, which restricts negative transfer in constraints of objective function. At the same time representative dataset in source domain are obtained to reduce the size of training data and improve training efficiency. The experimental results on several real-world datasets show the effectiveness of CMATLSVM, and it also has certain advantages compared with benchmark algorithms.