基于支持向量机的增量式算法精选.pdf

42 12 浙 江 大 学 学 报( ) Vol.42 o. 12 2008 12 Journal of Zhejiang University(Engineering Science) Dec. 2008 DOI: 10.3785/j.issn. 1008973X.2008. 12.015 1 2 1, 2 黄启春, 刘仰光 , 何钦铭 ( 1. , 310027; 2. , 315100) : , . ,. . , , . ,. ,. : ; ;; ; : T P181 : A : 1008973X(2008) Incremental learning algorithm based on support vector machine 1 2 1,2 HUA G Qichun , LIU Yangguang , HE Qinming (1. College of Comp uter Science and Technology , Zhej iang University, H angz hou 310027, China; 2. ingbo Institute Technology , Zhej iang University , ing bo 315100, China) Abstract: An incremental learning algorithm based on support vector machine was proposed to process large scale data or data generated in batches. Initial goal concept learned by standard support vector ma chine algorithm was updated by an updating model. T he model solved a convex quadratic programming similar to standard support vector machine algorithm. The algorithm proves that not only classification hy perplane but also the representation of classification hyperplane arenot changed during incremental learning procedure if the new samples arenot in the boundary in separable case. Decreasing learning procedure is easy to implement without extra computation and learning time is inversely proportional to incremental learning step compared with existing incremental learning support vector algorithms. Results show that the algorithm satisfies the three criteria of stability, improvement and recoverability. Key words: machine learning; pattern classification; support vector machine; incremental learning algorithm [ 12]