Unconstrained Minimization Methods for Solving Training Problems Arising in Neural Net Applications

解决神经网络应用中出现的训练问题的无约束最小化方法

• 批准号: 9000560
• 负责人: James Falk
• 金额: $ 3万
• 依托单位: George Washington University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1990
• 资助国家: 美国
• 起止时间: 1990-01-15 至 1991-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9000560&HistoricalAwards=false
• 关键词: Unconstrained; Minimization; Methods; Solving; Training

The P.I. will attempt to apply current methodology from nonlinear programming to the solution of unconstrained optimization problems arising in applications of neural nets. The four main areas of investigation are: solving large problems using recent developments in conjugate gradient and Newton methods; development of methods for obtaining global as opposed to strictly local solutions; application of sensitivity theory to provide post-optimality analysis; use of factorable function theory to develop varieties of functional forms describing the input-output relations.

P.I.将尝试将目前的方法从非线性编程应用于神经网应用中产生的无约束优化问题的解决方案。 调查的四个主要领域是：使用共轭梯度和牛顿方法的最新发展解决大问题；开发获得全球而不是严格本地解决方案的方法； 敏感性理论的应用以提供当前的分析；使用可分解功能理论来开发描述输入输出关系的功能形式的品种。