摘要
针对大规模界约束优化问题,列举了四种有效集识别策略,每次迭代它们允许多个有效约束的指标加到工作集或从工作集中去掉.在1998年Facchinei等人提出的有效集算法基础上,写出有效集拟牛顿算法(ASNA)框架用于测试不同的有效集识别策略.采用特殊的方法,由非线性无约束问题产生若干界约束极小化的测试问题,通过数值测试发现Facchinei等人同年提出的精确有效集识别函数不适用于本文的ASNA算法,最终分析了其余三种识别策略的优缺点.
We list four different active set identification techniques in this paper, which can add to or drop from the current estimated active sets many constraints at each iteration. It＇s possible to envisage these techniques suitable to solve large scale problems. We develop an active set quasi-Newton （ASNA） algorithm based on [4]. Numerical results show that the accurate active set identification techniques which was proposed by Facchinei in 1998 does not suit ASNA, at last the other three strategies are analyzed.
出处
《数值计算与计算机应用》
CSCD
北大核心
2009年第1期41-47,共7页
Journal on Numerical Methods and Computer Applications
基金
国家自然科学基金(10571109)资助项目.
关键词
有效集
界约束
大规模问题
支持向量机
active sets
bound constraints
large scale problems
support vector machine
