Incremental Parallel Support Vector Machines for Classifying Large-Scale Multi-class Image Datasets

In this paper, we propose an incremental parallel support vector machines (SVM) training with stochastic gradient descent (SGD) for dealing with the very large number of images and large-scale multiclass on standard personal computers (PCs). The two-class SVM-SGD algorithm is extended in several ways to develop the new incremental parallel multi-class SVM-SGD in large-scale classifications. We propose the balanced batch SGD of SVM (BBatch-SVM-SGD) for trainning two-class classifiers used in the one-versus-all strategy of the multi-class problems and the incremental training process of classifiers in parallel way on multi-core computers. The numerical test results on ImageNet datasets show that our algorithm is efficient compared to the state-of-the-art linear SVM classifiers in terms of training time, correctness and memory requirements.