Abstract:

The relationship between KKT conditions and the studied sample and the analysis of the change of support vector after the addition of incremental samples to the training set on the basis of an improved Upper Limit on Increment incremental learning algorithm. According to the KKT conditions for this algorithm the corresponding samples were divided into three categories: the RIG distributed outside the interval of classifier; the MAR at intervals on the classification and the ERR inside the intervals of classification. And ERR set after each training was preserved and combined with the incremental sample of the next training. Experiments show that the algorithm is feasible and effective.

Key words: support vector, incremental learning, support vector machine (SVM)