CAREER: New Algorithmic Approaches to Computationally Challenging Stochastic Supply Chain and Revenue Management Models

这个教师早期职业发展（Career）项目的研究目标是开发一个统一的计算和理论框架，以弥合多周期随机优化模型与供应链和收入管理中具有计算挑战性的实际大规模应用之间的差距。考虑像亚马逊这样的公司，它必须管理超过4000万种不同的商品，才能及时满足客户的购买需求。亚马逊必须做出许多日常决策，例如每种商品的库存数量，库存的位置以及如何将订单运送给客户。管理这么大的供应链是非常具有挑战性的，特别是因为未来的客户需求和供应是不确定的，并且随着时间的推移而波动。需求预测是管理未来不确定性的最有效工具之一。然而，如何利用需求预测来设计有效的库存控制政策，使供需相匹配，对研究者和实践者来说都是一个具有挑战性的问题。另一个例子是劳动力收入管理优化，像IBM这样的公司必须随着时间的推移管理技术工人池，以处理多个咨询项目，旨在选择最有利可图的项目。传统的建模工具，如动态规划，在研究这些模型的最优策略的结构特性方面是有效的。然而，他们通常不会。T导致有效的过程来计算实际实例的最优策略，甚至是好的策略。本研究项目旨在开发一种新的算法框架，在一般建模假设下研究这些问题，以捕获其实际方面。新算法将基于几种新技术，如边际成本核算方案和成本平衡技术，并提供易于实施但可证明接近最优的政策。一个关键的研究方法是使用理论和计算性能分析来指导新算法的发展。通过与行业伙伴的合作，这些算法将在真实数据上进行测试。该提案涉及供应链和收益管理中的几个广泛应用领域。如果成功，这项研究的结果将导致一个统一的建模和算法框架，以更广泛的视角研究这些实际问题，而不是目前的最新技术。这将扩展对这些具有挑战性问题的理论和计算理解。从长期来看，将捕捉问题的实际方面的更复杂的模型与概念上简单而有效的算法结合起来，可能会显著改善各自供应链和其他业务环境的性能和效率。与业界合作伙伴的合作将被用来提高这个研究项目的实际影响，并丰富学生的课堂体验。

The research objective of this Faculty Early Career Development (CAREER) project is the development of a unified computational and theoretical framework that will bridge the gap between multi-period stochastic optimization models and computationally challenging practical large-scale applications in supply-chain and revenue management. Consider a firm like Amazon that has to manage over 40 million different items to satisfy customer purchases in a timely manner. Amazon has to make many daily decisions, such as how many units of each item to stock, where to locate inventories, and how to ship orders to customers. The management of this large supply chain is very challenging especially because the future customer demands and supplies are uncertain and fluctuate over time. Demand forecasts are one of the most effective tools in managing future uncertainties. However, how to use demand forecasts to devise an effective inventory control policy that matches supply and demand is a challenging problem both for researchers and practitioners. Another example is workforce revenue management optimization, where companies like IBM have to manage pools of skilled workers over time to handle multiple consulting projects, aiming at choosing the most profitable ones. Traditional modeling tools like dynamic programming are effective in studying structural properties of the optimal policies of some of these models. However, they typically don?t lead to efficient procedures to compute optimal or even good policies for practical instances. This research project seeks to develop a new algorithmic framework to study these problems under general modeling assumptions that capture their practical aspects. The new algorithms will be based on several new techniques, such as marginal cost-accounting schemes and cost-balancing techniques, and provide simple to implement, yet provably near-optimal policies. A key research methodology is the use of theoretical and computational performance analysis to guide the development of the new algorithms. Through collaborations with industry partners the algorithms will be tested on real data.The proposal addresses several broad application domains in supply-chain and revenue management. If successful, the results of this research will lead to a unified modeling and algorithmic framework to study these practical problems in broader perspectives than the current state-of-the-art. This will expand the theoretical and computational understanding of these challenging problems. In the longer-term, the combination of more sophisticated models that capture the practical aspects of the problems together with conceptually simple and efficient algorithms is likely to lead to significant improvements in the performance and efficiencies of the respective supply chains and other business environments. Collaboration with industry partners will be used to enhance the practical impact of this research project, and to enrich the classroom experience for students.