【摘要】 提出一种快速的支撑向量回归算法。首先将支撑向量回归的带有两组约束的二次规划问题转化为两个小的分别带有一组约束的二次规划问题,而每一个小的二次规划问题又采用一种快速迭代算法求解,该迭代算法能从任何初始点快速收敛,避免了二次优化问题求解,因此能显著提高训练速度。在多个标准数据集上的实验表明,该算法比传统支撑向量机快很多,同时具有良好的泛化性能。更多还原

【Abstract】 This paper proposed a fast support vector regression algorithm.This algorithm converts the quadratic programming problems(Qpps) with pair groups of linear inequality constraints to two small size Qpps with only one group of linear inequality constraints.Each of the small size Qpps is solved by an iterative algorithm.The iterative algorithm converges from any starting point and does not need any quadratic optimization packages.Thus this algorithm is fast.The experimental results on several benchmark datasets demonstrate the effectiveness of the proposed algorithm.更多还原