CAREER:Synthesis of Search Procedures for Constraint Programs

提案0642906《Career：Constraint Programs Search Procedure for Constraint Programs》PI：Laurent D.Michelle of ConnecticutUniversity of ConnecticutAbstra本研究旨在从高级约束编程模型中合成约束程序(CP)。约束规划是解决复杂优化问题的一种方法，这些问题普遍存在于我们的社会中，对许多行业都是至关重要的，几乎影响到我们日常生活的方方面面。然而，目前的约束编程方法需要大量的专业知识来设计和编程搜索程序，因此还没有被优化建模者广泛使用。这与数学规划的建模语言不同，后者现在是高度自动化的。这个项目研究从高级模型自动合成约束程序。它的方法利用约束程序在单个约束中表达组合结构的能力来推导出整个模型的对称性或支配性等关键性质。根据这些性质，它可以重新制定和综合适合于基本求解器技术的搜索过程，例如约束编程、局部搜索、整数规划，甚至混合。为了培养下一代工程师，PI正在开发针对高中数学和科学教师的材料，并将通过康涅狄格大学的达芬奇倡议传播这些努力。该项目的更广泛影响包括基于编程语言、编译器和人工智能的最新进展开发新一代建模语言和系统；这些语言和系统将能够服务于更广泛的优化建模者。

Proposal 0642906"CAREER: Synthesis of Search Procedures for Constraint Programs"PI: Laurent D. MichelUniversity of ConnecticutAbstractThis research aims at synthesizing constraint programs (CP) from high-level constraint programming models. Constraint programming is a method for solving complex optimization applications, which are ubiquitous in our society, are critical to many industries, and affect almost every aspect of our daily lives. However, current constraint programming methodologies require significant expertise to design and program search procedures, and consequently, are not yet widely used by optimization modelers. This is in contrast with modeling languages for mathematical programming, which are now highly automated.This project investigates the automatic synthesis of constraint programs from high-level models. Its approach exploits the ability of constraint programs to express combinatorial structures in individual constraints to derive key properties like symmetries or dominance for entire models. From these properties, it can reformulate and synthesize search procedures adapted to the underlying solver technology, such as constraint programming, local search, integer programming, or even hybrids. To educate the next generation of engineers, the PI is developing material targeting high school math and science teachers and will disseminate these efforts via the DaVinci initiative at the University of Connecticut. The Broader Impacts of this project include the development of a new generation of modeling languages and systems based on recent advances in programming languages, compilers and artificial intelligence; these will be able to serve a wider audience of optimization modelers.