An active learning method of a support vector machine

Abstract

The invention provides an active learning method of a support vector machine. Selecting as far as possible samples from an initial training set by using a maximum and minimum distance method as an initial clustering center of the K-means algorithm can avoid the problem that a clustering center is too close when an initial value of the K-means algorithm is selected, meanwhile, a distance thresholdvalue is set in the iteration process to remove a part of sample points far away from a current optimal classification hyperplane, and K-is applied to the remaining sample points. The means algorithmis used for clustering to determine the optimal classification hyperplane capable of meeting the index, and the efficiency of dividing the initial sample set can be improved.

Description

technical field [0001] The invention relates to the technical field of computers, in particular to an active learning method of a support vector machine. Background technique [0002] Active learning algorithms such as figure 1 As shown), the main purpose is to use the existing knowledge and use the least training data to obtain better performance classification results. Compared with traditional passive learning, active learning selects the most valuable samples for labeling in each iteration, which reduces the time for labeling meaningless sample points. [0003] In the training phase of the active learning algorithm, since a small number of training sample points that are most likely to become support vectors are actively added and marked, the training efficiency of the support vector machine is greatly improved. The K-means algorithm is a classic clustering algorithm that uses distance as a similarity evaluation index. This algorithm has the advantages of simplicity, n...

AI Technical Summary

Problems solved by technology

[0004] However, in the active learning process based on the K-means algorithm, the clustering data K in the K-means algorithm needs to be given in advance, and the support vector machine is a classification problem for the second class, that is, K=2. In addition, the K-means algorithm The degree of dependence on the selection of the initial value is relatively large

And at present, when selecting the initial sample points, it is always randomly or according to the prior probability, and the distribution of the prior probability does not have a good measure, which will cause the classifier to have a relatively large number of iterations in the initial stage of training. , which reduces the classification efficiency; at the same time, when selecting the most valuable samples, because the sample points with the greatest uncertainty are selected, and these points are often distributed in the critical area, the algorithm itself is susceptible to the influence of singular points

Images