MDTMM-LS: A New Matrix-Based Classification Method Based on Multidomain Information Sharing

The traditional matrix classifier does not perform well in dealing with the problems of limited information and low computational efficiency, especially in the multi-task environment. For this reason, a new classification method is proposed, called multi-domain twin matrix machine with least squares (MDTMM-LS). In MDTMM-LS, the design of the common domain kernel (CDK) realizes multi-objective domain learning while making full use of the structural information of matrix samples, and avoids the iterative operations as in the traditional matrix-based classifiers. Meanwhile, in the process of constructing multi-objective domain models, the least squares loss is used as a constraint, which can efficiently avoid inverse matrix calculations, thereby reducing computational complexity. Through the experimental verification of two groups of rolling bearing and gear related data, the results show that MDTMM-LS has significant advantages in multi-target domain processing.