摘要
传统的二次规划由于涉及大量的矩阵运算,运算速度慢成为支持向量机的最大缺点.已有的乘性规则仅适于非负二次凸规划问题,推导出了求解支持向量机中混合约束二次凸规划的乘性规则,利用这一乘性规则极大地提高了优化速度.该方法提供了一种直接优化的方法,其所有变量可以并行迭代,乘性规则可以使得二次规划的目标函数单调下降到它的全局最小点.仿真试验结果表明了该算法有效性.
Due to intensive matrix computation, the speed of the quadratic equation remains slow. The multiplicative updates available are only suited for nonnegative quadratic convex programming. In this article, the multiplicative updates are derived for mixed constraint optimizations, dramatically speeding up optimization rate. This method provides an extremely straightforward way to implement support vector machines (SVMs) where all the variables can be iterated in parallel. The multiplicative updates converge to global minimum point by monotonically reducing the target function of quadratic programming. Experimental results have confirmed the effectiveness of our approach.
出处
《智能系统学报》
2007年第2期74-77,共4页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金资助项目(60574075)
关键词
支持向量机
二次凸规划
混合约束
乘性规则
support vector machine
quadratic convex programming
mixed constraint
multiplicative update