SVM Incremental Learning with Model Adjusting in the Wise-Increasing Neighborhood
This paper presents an Incremental Learning algorithm for SVM (ILSVM) on observing a new data. ILSVM is characterized by locally adjusting model in an informative neighborhood of the new data. Although the objective for model adjusting is optimized locally, its definition involves all data information, thus fostering the generalization for the result of local learning. The neighborhood where optimization takes place is formulated under the guidance of two instructive metrics that represent directions along which the truly concerned neighbors can be found. These two metrics are derived from SVM decision interface and their employment in ILSVM assists probing a desired neighborhood at a fast speed. Hyper parameters of SVM are learned from data context, to bring adaptation and avoid pricy cost. Experiments on real datasets evidence the performance and efficiency of ILSVM over the state-of-the-art methods.
Ping Ling Chunguang Zhou Zhe Wang
College of Computer Science and Technology Jilin University Key Laboratory for Symbol Computation an College of Computer Science and Technology Jilin University Key Laboratory for Symbol Computation an College of Computer Science and Technology Jilin University Key Laboratory for Symbol Computation an
国际会议
南宁
英文
2007-07-20（万方平台首次上网日期，不代表论文的发表时间）