Fast SVM Incremental Learning Based on the Convex Hulls Algorithm

Abstract

To reduce the computational cost of the incremental learning, a fast SVM incremental learning algorithm based on the convex hulls algorithm is proposed in this paper. The given algorithm is based on utilizing the result of the previous training effectively and retaining the most important samples for the incremental learning to reduce the computational cost. In the process of incremental learning, the convex hull vectors of the previous training and the newly added samples constitute the current training sample set, the current training sample set is pre-extracted from the geometric point of view by using the convex hulls algorithm, the central distance ratio method is used to obtain the between-class convex hull vectors, and the between-class convex hull vectors are used as the training samples in the SVM incremental training. Experiments prove that the given algorithm has better classification performance.