Auto-weighted support vector machines for training sets with multiduplicate samples

By analyzing C-SVM theoretically and experimentally, we found that it was over-dependent on each training sample, even if the samples are multiduplicate. This dependence would result in more time for training and more support vectors. More support vectors result in more time for decision. In order to overcome this problem, we propose an extended C-SVM. termed auto-weighted support vector machine. Auto-weighted support vector machine multiplies each slack variable by a weight factor and automatically increases the weight factor by the times of duplication during the phase of reading training samples. Experiments showed that auto-weighted support vector was faster than C-SVM in both training and decision if the training sets had multiduplicate samples.