Online Incremental and Decremental SVDD Learning Based on CP

Abstract

When there is a new sample, the traditional training algorithm will train all the samples, which costs a lot of time, and can’t meet requirement of real-time in online learning. In order to solve this problem, an incremental SVDD learning algorithm based on CP is presented, and the calculation and derivation process is presented in details. The new algorithm can make full use of the existing training results, and update learning on the original model according to the new sample, which can not only inherit the previous accumulated knowledge information, but also greatly reduce the training time, and which can true realize the online learning. Experiments show that the algorithm trains fast and the classification accuracy of it is high, which has value in engineering application.