Collaborative Research: Coordinating Offline Resource Allocation Decisions and Real-Time Operational Policies in Online Retail with Performance Guarantees

许多行业必须间歇性地做出战术资源分配决策，这限制了使用这些资源的后续操作决策的有效性。例如，在线零售商必须决定如何在地理分布的仓库中分配可用库存，从而影响单个订单履行决策。在目前的实践中，资源分配决策没有考虑到操作决策的性质，这可能导致效率的重大损失。本研究计划将为国家的经济福利作出贡献，为这些问题制定一套总体策略，可针对不同的问题设置进行定制，提供一套适用于许多行业的独立于环境的协调机制。教育和推广活动将让学生参与实践优化项目，创建数据中心决策的新教育计划，并通过从协调资源分配决策和实时操作的角度来解决城市地区的食物沙漠问题。目前的知识状况并没有提供有效的算法来协调战术资源分配决策和使用这些资源的作战决策。造成这种差距的主要原因有两个：计算最优运营策略往往需要求解高维动态规划，最优价值函数可能缺乏结构，无法为资源分配决策提供明确的依据。这个项目将为独立于应用程序设置的协调建立一个通用的近似框架。该框架使用代理函数来开发最优操作策略性能的上界和近似操作策略性能的下界。限定两个代理之间的相对差距将为协调问题提供性能保证。选择适当的代理函数将允许将近似框架应用于不同的应用，提供具有可证明的性能保证的有效近似解。为了实现这样的解决方案，将结合组合优化和动态优化以及离散选择和价格响应建模的思想。生成的算法将在几个公开可用的数据集上进行测试。该奖项反映了美国国家科学基金会的法定使命，并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，被认为值得支持。

Many industries must intermittently make tactical resource allocation decisions that constrain the effectiveness of subsequent operational decisions using these resources. For example, online retailers must decide how to distribute their available inventory among geographically distributed warehouses, influencing individual order fulfillment decisions. In current practice the resource allocation decisions do not consider the nature of the operational decisions, which can result in significant losses in efficiency. This research project will contribute to the economic welfare of the nation by developing a general strategy for these problems that can be customized to different problem settings, providing a context-independent coordination mechanism that can be applied to many industries. The educational and outreach activities will involve students in hands-on optimization projects, create new education programs in datacentric decision-making, and address the food deserts problem in urban areas by viewing the problem from the lens of coordinating resource allocation decisions with real-time operations.The current state of knowledge does not provide efficient algorithms for coordinating tactical resource allocation decisions with the operational decisions that use those resources. There are two main reasons for this gap: computing the optimal operational policy often requires solving a high-dimensional dynamic program, and the optimal value functions may lack structure, providing no clear basis for resource allocation decisions. This project will build a general approximation framework for coordination that is independent of the application setting. The framework uses surrogate functions to develop an upper bound on the performance of the optimal operational policy and a lower bound on that of an approximate operational policy. Bounding the relative gap between the two surrogates will yield a performance guarantee for the coordination problem. Selection of appropriate surrogate functions will allow application of the approximation framework to different applications, providing efficient approximate solutions with provable performance guarantees. To achieve such solutions, ideas from combinatorial and dynamic optimization, as well as discrete choice and price response modeling will be combined. The resulting algorithms will be tested on several publicly available datasets.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.