Kernel Support Vector Machine for Domain Adaptation
领域适应核支持向量机

Domain adaptation learning is a novel effective technique to address pattern classification, in which the prior information for training a learning model is unavailable or insufficient. To minimize the distribution discrepancy between the source domain and target domain is one of the key factors. However, domain adaptation learning may not work well when only considering to minimize the distribution mean discrepancy between source domain and target domain. In the paper, we design a novel domain adaptation learning method based on structure risk minimization model, called DAKSVM (kernel support vector machine for domain adaptation) with respect to support vector machine (SVM) and least-square DAKSVM (LSDAKSVM) with respect to least-square SVM (LS-SVM), respectively to effectively minimize both the distribution mean discrepancy and the distribution scatter discrepancy between source domain and target domain in some reproduced kernel Hilbert space, which is then used to improve the classification performance. Experimental results on artificial and real world problems show the superior or comparable effectiveness of the proposed approach compared to related approaches.