υ-structured support vector machines

This paper considers a v-structured support vector machine (v-SSVM) which is a structured support vector machine (SSVM) incorporating an intuitive balance parameter v. In the absence of the parameter v, cumbersome validation would be required in choosing the balance parameter. We theoretically prove that the parameter v asymptotically converges to both the empirical risk of margin errors and the empirical risk of support vectors. The stochastic subgradient descent is used to solve the optimization problem of the v-SSVM in the primal domain, since it is simple, memory efficient, and fast to converge. We verify the properties of the v-SSVM experimentally in the task of sequential labeling handwritten characters.