Collaborative Research: PPoSS: Large: A Full-stack Approach to Declarative Analytics at Scale

该项目研究了表达性编程系统的全栈实现方法，这些方法有效地弥合了人类级别规范与大规模复杂推理任务的高性能实现之间的差距。声明性语言允许程序员提供高级规则和声明，这些规则和声明将一些寻求的解决方案定义为要由计算机自动实现的潜在含义。该项目的新颖之处在于将这种高性能声明性推理的愿景扩展到结构化，高阶和概率公式以及下一代超级计算机和基于云的集群。该项目对关键应用领域的应用程序设计师和程序员产生了影响，包括精准医学、随机建模、软件验证、图形分析和安全性。该项目正在开发能够为跨学科用户实现真正可扩展推理的开源工具、编程语言和框架。下一代艾级系统的复杂性带来了关键挑战：管理增加的并行性、异构性、图形处理单元（GPU）、深层内存层次结构以及整个软件堆栈的性能调优。随着即将到来的高性能计算系统的硬件配置的复杂性和多样性的增加，手工编写可维护和可扩展的应用程序变得困难。现代链式正向推理系统正在扩展结构化的高阶数据、概率语义、格序、递归聚合和一阶理论，这带来了关键的实现挑战-特别是在并行环境中。在这个项目中，研究人员正在开发一个统一的、可调的、全栈的基础，用于大规模部署高度表达性的链式向前编程。这个奖项反映了NSF的法定使命，并被认为值得通过使用基金会的知识价值和更广泛的影响审查标准进行评估来支持。

The project investigates full-stack implementation methodologies for expressive programming systems that effectively bridge the gap between human-level specification and high-performance implementation of complex reasoning tasks at scale. Declarative languages permit a programmer to provide high-level rules and declarations that define some sought-after solution as a latent implication to be materialized automatically by the computer. The project's novelties are to scale this vision of high-performance declarative reasoning both to structured, higher-order, and probabilistic formulations and to the next generation of supercomputers and cloud-based clusters. The project's impacts are on application designers and programmers in key application areas, including precision medicine, stochastic modeling, software verification, graph analytics, and security. The project is developing open-source tools, programming languages, and frameworks capable of enabling truly scalable reasoning for users across disciplines.The complexities of next-generation exascale systems pose key challenges: managing increased parallelism, heterogeneity, graphic processing units (GPUs), deep memory hierarchies, and performance tuning across the full software stack. With this increasing complexity and diversity in the hardware configuration of upcoming high-performance computing systems, it becomes difficult to write maintainable and scalable applications by hand. Modern chain-forward reasoning systems are being extended with structured, higher-order data, probabilistic semantics, lattice orderings, recursive aggregation, and first-order theories, posing key implementation challenges - especially in a parallel setting. In this project, the investigators are developing a unified, and tunable, full-stack foundation for highly expressive chain-forward programming to be deployed at scale.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.