A novel DoubleMinOver classifier based On second-order tensor

It is well-known that the DoubleMinOver classifier as an extendibility of MinOver was developed to provide a maximum margin solution with a bias. But, it can be found that the DoubleMinOver algorithm only classifies the vector pattern and fails to work in the pattern represented with the second-order tensor. In this paper, we propose a novel DoubleMinOver classifier named STDoubleMinOver that can directly classify the second-order tensor pattern. Compared with the existing DoubleMinOver with only one weight w, the proposed STDoubleMinOver induces two weight vectors u and v so as to deal with the second-order tensor. The experiments here have demonstrated that the proposed STDoubleMinOver has a superior classification to the vectorized DoubleMinOver.