Accelerating generalized iterative scaling based on staggered aitken method for on-line conditional random fields

In this paper, a convergent method based on Generalized Iterative Scaling (GIS) with staggered Aitken acceleration is proposed to estimate the parameters for an on-line Conditional Random Field (CRF). The staggered Aitken acceleration method, which alternates between the acceleration and non-acceleration steps, ensures computational simplicity when analyzing incomplete data. The proposed method has the following advantages: (1) It can approximate parameters close to the empirical optimum in a single pass through the training examples; (2) It can reduce the computing time by approximating the Jacobian matrix of the mapping function and estimating the relation between the Jacobian and Hessian in order to replace the inverse of the objective function's Hessian matrix. We show the convergence of the penalized GIS based on the staggered Aitken acceleration method, compare its speed of convergence with those of other stochastic optimization methods, and illustrate experimental results with two public datasets.