CTJLSVM: Componentwise Triple Jump Acceleration for Training Linear SVM

The triple jump extrapolation method is an effective approximation of Aitken’s acceleration for accelerating the convergence of many machine learning algorithms that can be formulated as fixedpoint iteration. In the remainder of this abstract, we briefly review the general idea of the triple jump method and then describe how to apply it to accelerate stochastic gradient descent (SGD) for training linear support vector machines (SVM).