摘要
现有的大多数分类问题都能转化成一个正定二次规划问题的求解.通过引入滤子方法,并结合求解非线性规划的原始对偶内点法,给出求解正定二次规划的滤子内点算法.该算法避免了使用效益函数时选取罚因子的困难,在较弱的假设条件下,算法具有全局收敛性.
Most existed classification problems can be converted into a positive definite quadratic program. In this paper, a filter interior-point algorithm for solving positive definite quadratic pregram is proposed by combining the filter technique with the primal-dual interiorpoint methods for NLP. The new algorithm can avoid the difficulties in the use of a penalty functon. Under mild assumptions, it is proved to be globally convergent.
出处
《数学的实践与认识》
CSCD
北大核心
2008年第21期127-133,共7页
Mathematics in Practice and Theory
基金
国家自然科学基金资助项目(10571109)
山东省科技厅(2006GG3210009)
山东省教育厅资助项目(J06P14)
关键词
分类问题
支持向量机
正定二次规划
滤子
原始对偶内点法
全局收敛性
classification problems
support vector machine
positive definite quadraticprogram
filter
primal-dual interior-point methods
global convergence
