Generalized multi-view learning based on generalized eigenvalues proximal support vector machines

Multi-view learning based on generalized eigenvalue proximal support vector machines has brought enormous success by mining the consistency information of two views. Nevertheless, it only aims to handle two-view cases and cannot handle generalized multi-view learning cases (above two views). It also omits the complementarity information among views. In this paper, two generalized multi-view extensions of generalized eigenvalue proximal support vector machines are presented which take advantage of the multi-view co-regularization term to mine the consistency information and the weighted value to mine complementarity information. Experimental results performed on synthetic and real world datasets demonstrate that they can provide higher performance than the relevant two-view classification algorithms.

Automated summary

This paper proposes two generalized multi-view extensions of generalized eigenvalue proximal support vector machines, which utilize multi-view co-regularization term and weighted value to mine consistency and complementarity information. Experimental results demonstrate that these methods outperform relevant two-view classification algorithms in terms of performance.