Multi-view learning with enhanced multi-weight vector projection support vector machine.

Multi-view learning aims on learning from the data represented by multiple distinct feature sets. Various multi-view support vector machine methods have been successfully applied to classification tasks. However, the existed methods often face the problems of long processing time or weak generalization on some complex datasets. In this paper, two multi-view enhanced multi-weight vector projection support vector machine models are proposed. One is a ratio form of multi-view enhanced multi-weight vector projection support vector machine (R-MvEMV), while the other is a difference form (D-MvEMV). Instead of searching for specific classification hyperplanes, each proposed model tries to generate two projection matrices composed of a set of projection vectors for each view. A co-regularization term is added to maximize the consistency of different views. R-MvEMV and D-MvEMV can be simplified to two generalized eigenvalue problems and two eigenvalue problems, respectively. The optimal weight vector projections are the eigenvectors corresponding to the smallest eigenvalues. Some numerical tests are conducted to compare the proposed methods with the other state-of-art multi-view support vector machine methods. The numerical results show the better classification performance and higher efficiency of the proposed methods.