CAJ | 학술논문

求解支持向量机大规模分类问题时,系数矩阵的存储和计算是非常困难的.借助分解技术,把问题分解成多个维数较低的二次规划问题.利用增广拉格朗日函数将子问题转化成只含有界约束的形式,再用修正子空间有限记忆BFGS方法解子问题,节省了存储空间,提高了求解效率.
구해지지향량궤대규모분류문제시,계수구진적존저화계산시비상곤난적.차조분해기술,파문제분해성다개유수교저적이차규화문제.이용증엄랍격랑일함수장자문제전화성지함유계약속적형식,재용수정자공간유한기억BFGS방법해자문제,절성료존저공간,제고료구해효솔.
The storage and computing of coefficient matrix were very difficult for solving large-scale SVMs.At each decomposition iteration,the problem was split up into smaller quadratic programming subproblems.The augmented lagrangian scheme was used to transform the subproblems into problems only with a set of bound constrains,then the subproblems were solved by modified subspace limited memory BFGS algorithm.This method could save storage space considerably and improve the efficiency.