Training Data Selection for Support Vector Machines

摘要: In recent years, support vector machines (SVMs) have become a popular tool for pattern recognition and machine learning. Training SVM involves solving constrained quadratic programming problem, which requires large memory enormous amounts of training time large-scale problems. contrast, the decision function is fully determined by small subset data, called vectors. Therefore, it desirable to remove from set data that irrelevant final function. this paper we propose two new methods select training. Using real-world datasets, compare effectiveness proposed selection strategies in terms their ability reduce size while maintaining generalization performance resulting classifiers. Our experimental results show significant amount can be removed our without degrading