CAREER: The Effects of Centralized and Decentralized Sequential Decisions on System Performance

点击翻译按钮获取中文摘要

The goal of this Faculty Early Career Development (CAREER) grant is to develop a comprehensive understanding of the immediate and long-term effects of centralized and decentralized sequential decisions. Individuals, organizations and firms make decisions under uncertainty on a regular basis. Such decisions are often complex and difficult to make, and their implications are difficult to measure. This research will develop approaches that quantify the uncertainty arising during the decision-making process. In addition, this research will also address decisions that involve sequential routing and scheduling and require the completion of multiple tasks in an interrelated network. The findings will highlight the effects of multiple sequential decisions that go beyond (nearly-) optimizing different complex systems. Such sequential decision models provide a natural framework for the analysis of inventory control, queueing, revenue management, routing, and many other problems frequently arising in practice. Therefore, this research will provide new fundamental theoretical results for academics as well as key insights for practitioners. Furthermore, this award will foster the education and the development of undergraduate and graduate students who will take an active part in several research projects. This award will also help expanding the graduate curriculum through the development of new Ph.D. courses that integrate the results of this research. In all the activities, the principal investigator seeks to promote the participation of underrepresented groups in the sciences and engineering.The award will support two complementary lines of work. The first considers centralized decisions. It is motivated by the a priori possibility that the realized rewards may often be far away from their optimal expected value, and it provides a detailed probabilistic analysis of dynamic programming policies. Specific topics include limit theorems for controlled transient systems (including knapsack problems), distributional equivalences across formulations with different horizon lengths, and the identification of policies that lead to the same limiting distribution as the one obtained by implementing the optimal policy. The second line of work studies decentralized decisions. It tackles the complexity that comes from combining a network structure with the routing of self-interested parties who require multiple different services. As a result, this research will characterize equilibrium routing behaviors in queuing networks, compare the performance of a system with self-interested agents with the performance of the same system when managed by a central planner, and design mechanisms that aim to align the centralized and the decentralized solutions.