ALGORITHMS: Multi-Algorithm Parallel Optimization of Costly Functions

算法：高代价函数的多算法并行优化

• 批准号: 0305583
• 负责人: Christine Shoemaker
• 金额: $ 38万
• 依托单位: Cornell University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2003
• 资助国家: 美国
• 起止时间: 2003-07-01 至 2007-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0305583&HistoricalAwards=false
• 关键词: ALGORITHMS; Multi; Algorithm; Parallel; Optimization

The proposed procedure MAPO(Multi-Algorithm Parallel Optimization) iteratively uses a committee of serial response surface algorithms to facilitate the generation of a sizeable number of distinctly different points for costly function evaluation in parallel. We refer to the methods as algorithm-experts. This approach can be effective because the computation time for one serial response surface optimization algorithm is very small in comparison to the CPU time required to evaluate costly f(x). Hence MAPO has a committee of experts, each one of which selects several candidate points x for costly function evaluation.The two main classes of algorithm-experts use Radial Basis Functions and Neural Nets. The algorithm will be applied to a range of difficult test problems and to three classes of costly real engineering functions. Two of these applications come from the PI's own research projects on environmental pollution and safety of drinking water. Another project involves a finite element model of a mechanics system based on partial differential equations provided by the NSF ITR Adaptive Software Project at the Cornell Theory Center.

建议的程序MAPO（多算法并行优化）迭代地使用一个委员会的串行响应面算法，以便于产生大量的明显不同的点，并行昂贵的功能评估。 我们将这些方法称为算法专家。这种方法可以是有效的，因为一个串行响应面优化算法的计算时间是非常小的相比，所需的CPU时间来评估昂贵的f（x）。 因此，MAPO有一个专家委员会，每个专家委员会选择几个候选点x进行代价函数评估。两类主要的算法专家使用径向基函数和神经网络。该算法将被应用到一系列困难的测试问题和三类昂贵的真实的工程功能。 其中两项申请来自PI自己的环境污染和饮用水安全研究项目。另一个项目涉及康奈尔理论中心NSF ITR自适应软件项目提供的基于偏微分方程的力学系统有限元模型。