Support Vector Machine Training Algorithm And Its Improvement

Support Vector Machines (SVM) are a kind of novel machine learning methods which have become the hotspot of machine learning because of their excellent learning performance. They also have successful applications in many fields, such as: face detection, handwriting digit recognition, text auto-categorization, etc. But as a new technique, SVM also have many shortcomings that need to be researched, such as: it costs too much time when training large-scale data set and its alogrithm is so difficult to implentmenion.In this paper, we main discuss the training alogrithm of SVM. At first, this paper introduced the basic concept of SVM theory. Then we discussed the history of the training alogrithm of SVM. As following, we paid main attention to the two successful training alogrithms - SVMlight and SMO. Both of the two alogrithms improved the trianing speed on large-scale training set greatly. Especially for SVMlight, it use shrinking and kernel caching to reduce the total computational quantity, so it speedup the trainging.For large-scale training set, to fasten the training speed, this paper combined the advanteges of SVMlight and SMO, and used some information of experiments, finally proposed a improved SVM training alogrithm: Set-Division-Based SMO(SSMO). According to the violation degree of sample, we can divide the training set to servral sets. After each iteration, SSMO makes use of the sets to update the working set and relevent parameters. So the cost of joint optimazation in each iteration reduced. Finally, SSMO can speedup the training.At last, this paper us two data sets to test SSMO alogrithm. The result is that SSMO can speedup the training procedure greatly. SSMO espescially fits for the training of largr-scale training set.