High-Performing Training Large-Scale Binary Nonlinear SVMs Using Mean Decision Rule Method with Smart Sampling

In spite of multiple investigations in the field of large-scale SVM tasks, the problem of their solution is still actual. Previously, we proposed the kernel-based mean decision rules method with smart sampling (SS-KMDR). According to experiments, it provides the possibility to obtain results close in quality to the reference ones, but for much less time compared to traditional methods. In this paper, we present a new high-performance implementation (HP-SS-KMDR), which allows us to further speed-up computations through the development of efficient parallel algorithms and modern technologies for parallel and distributed computing. Besides, experiments show that in a number of cases the proposed HP-SS-KMDR method allows us to increase the recognition quality compared to the sequential version due to the possibility to process more data in less time.