Abstract: To solve the problem that the optimization of Least Square Support Vector Machine (LSSVM) consumes too much time, a new method called Newton-LSSVM, which used Newton optimization to solve the optimization problem of LSSVM, was proposed. The method first converted LSSVM to unconstrained optimization, and then used Newton optimization to solve the optimization problem iteratively. The experimental results show that Newton-LSSVM can reduce the training time greatly but do not decrease the generating ability of LSSVM.

Key words: supervised learning, Least Square Support Vector Machine (LSSVM), optimization algorithm

摘要： 为了解决最小二乘支撑向量机（LSSVM）优化问题需要耗费大量时间的问题，提出了利用牛顿优化法来解决LSSVM优化问题的方法（称为Newton-LSSVM）。首先把LSSVM优化问题转化为无约束化优化问题的形式，然后再采用牛顿优化法来迭代求解。实验结果表明，该方法在大幅度减少LSSVM算法的训练时间开销的同时，能够获得与采用传统优化方式求解LSSVM优化问题一样的泛化能力。

关键词: 监督学习, 最小二乘支撑向量机, 优化算法