Research Of Online Learning Algorithms Based On Support Vector Machine

SVM(Support Vector Machine) theory has been recently drawed much more attention as an important research branch in Machine Learning technologies. Its theory is becoming a hot spot in research of Machine Learning.One of the characteristics of Support Vector Machine is pursuiting the target of minimum structural venture. And it is this characteristic that makes Support Vector Machine theory be widely accepted by its application value. As Support Vector Machine developed from theory research to application in industry, it is required to have more advanced ability. One of them is supporting On-line learning ability.On-line Learning algorithm is also another important research branch in Machine Learning. Due to the excellent theory value and application value of Support Vector Machine, its on-line learning algorithm has become another hot spot in Machine Learning. Several kinds of software algorithm of on-line learning of Support Vector Machine have been proposed in the recent years, and these algorithms has received great effects in applications. As the machine learning develops, the on-line algorithm of learning of Support Vector Machine is revealing its theorical value and applicational value. But very few algorithm of on-line learning of Support Vector Machine is proposed these years and, at the same time, seldom algorithm deserves the concept of "on-line". As a result, there is great value and developing space in this research branch.the research that can solve the above problems in this paper is listed as follows:1. A brief introduction to develop background of Support Vector Machine, several application field and theory of Support Vector Machine, and a detail introduction to the mathematical theory of Support Vector Machine, including deduction theory.2. Introduction and Summary of state-of-art and software implementation methods of algorithm of Support Vector Machine, and online learning algorithm of Support Vector Machine and the corresponding implementation thesis, and the summary of several online learning methods.3. For incremental learning, the shell vectors and the central density based on the support vector machine algorithm is proposed. The algorithm uses the shell vector and center density algorithm to pick out the potential original vector to transform the support vector, narrowing the number of samples to participate in the training. Through experiments shows that the algorithm can not only accelerate the training speed but also improve the classification accuracy; For the same amount of learning, based on sample window to select algorithm is proposed.The algorithm will make new vector be added to the sample with the sample sort, select retention some vectors that have higher training classification accuracy, improving the classification precision and keep the training time does not increase.