On coresets for support vector machines

We present an efficient coreset construction algorithm for large-scale Support Vector Machine (SVM) training in Big Data and streaming applications. A is a small, representative subset of the original data points such that model trained on provably competitive with set. Since size generally much smaller than set, our preprocess-then-train scheme has potential to lead significant speedups when SVM models. prove lower upper bounds required obtain small summaries problem. As corollary, we show can be used extend applicability any off-the-shelf solver streaming, distributed, dynamic settings. evaluate performance real-world synthetic sets. Our experimental results reaffirm favorable theoretical properties demonstrate its practical effectiveness accelerating training.