Abstract: According to the geometry distribution property of Support Vector(SV), this paper proposes the concept of adjacent boundary model and SV pre-selecting algorithm. By pre-selecting adjacent boundary samples, lots of samples are avoided to solve Quadratic Programming(QP) problems, which provides efficient training sets for SVM. Experiments show that it can improve efficiencies of training time and space by using the algorithm in LIBSVM.

Key words: adjacent boundary model, Support Vector Machine(SVM), active learning

摘要： 根据支持向量的几何分布特性，提出相邻边界模型的概念以及一种支持向量预选算法。该算法通过预选出相互邻近的边界样本，避免大量样本参与二次规划问题的求解，为支持向量机提供高效的训练集。实验结果证明，采用该预选算法的LIBSVM可以较大地提高训练的时间效率和空间效率。

关键词: 相邻边界模型, 支持向量机, 主动学习

CLC Number: 

• TP301.6