Collaborative Proposal: Resource Allocation with Learning in Dynamic and Partially Observable Networks

协作提案：动态和部分可观察网络中的资源分配和学习

• 批准号: 1538860
• 负责人: Pinar Keskinocak
• 金额: $ 18.88万
• 依托单位: Georgia Tech Research Corporation
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2015
• 资助国家: 美国
• 起止时间: 2015-08-01 至 2019-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1538860&HistoricalAwards=false
• 关键词: Collaborative; Proposal; Resource; Allocation; Learning

This collaborative project will research restoration of connectivity of interdependent road and service networks under uncertainty and learning. Real-life applications include post-disaster debris clearance to enable disaster response activities and interdependent infrastructure recovery and repair after a disruption. For example, when the road network is (partially) disrupted, edges (roads) need to be restored (e.g., through repair and debris clearance). The goal is to establish connectivity between supply and demand nodes in a timely manner to satisfy demand. Under uncertainty (about network conditions, and supply-demand levels), decision-making can be improved by collecting situational spatial data. However, data collection consumes time and resources. Furthermore, a common set of resources may perform both restoration and learning activities. Hence, under resource and time constraints, decisions on dynamically prioritizing edge recovery and resource allocation between recovery and learning are crucial for operational efficiency and effectiveness. This project considers the trade-off between learning and restoration activities, and decisions on (equitable and timely) resource allocation and frequency of information updates. The project will leverage the outreach network established by the Center for Health and Humanitarian Systems at Georgia Tech to disseminate results and interact with practitioners during project execution.Previous research on network connectivity in stochastic networks with learning is limited. If successful, this project will introduce network repair models that are characterized by limited resources, uncertainty, ability to reduce uncertainty by deploying resources, and the need to maintain fairness in service access. It will contribute to a deeper understanding of how to solve dynamic multi-stage decision problems in stochastic networks that offer an opportunity to update information and in which learning and recovery actions share the same resources. This research will lead to efficient solution approaches for finding optimal or near-optimal solutions. The research team will perform structural analysis of underlying problems on simple networks, evaluate different information update mechanisms, and use those to derive insights and generate customized solutions.

本合作项目将研究在不确定性和学习条件下相互依赖的道路和服务网络的连通性恢复。现实生活中的应用包括灾后碎片清理，使灾害响应活动和相互依赖的基础设施在中断后恢复和修复成为可能。例如，当道路网络（部分）中断时，需要恢复边缘（道路）（例如，通过维修和清除碎片）。目标是在供需节点之间及时建立连接，以满足需求。在不确定的情况下（关于网络条件和供需水平），可以通过收集情景空间数据来改进决策。但是，收集数据需要耗费大量的时间和资源。此外，一组公共资源可以同时执行恢复和学习活动。因此，在资源和时间的限制下，边缘恢复的动态优先级以及恢复和学习之间的资源分配决策对操作效率和有效性至关重要。该项目考虑了学习和恢复活动之间的权衡，以及（公平和及时的）资源分配和信息更新频率的决定。该项目将利用佐治亚理工学院卫生与人道主义系统中心建立的外展网络，在项目执行期间传播结果并与从业人员互动。以往对具有学习特性的随机网络的连通性研究比较有限。如果成功，该项目将引入网络修复模型，其特点是资源有限、不确定性、通过部署资源来减少不确定性的能力，以及需要保持服务访问的公平性。这将有助于更深入地理解如何解决随机网络中的动态多阶段决策问题，这些问题提供了更新信息的机会，并且学习和恢复行动共享相同的资源。这项研究将导致寻找最优或近最优解的有效解方法。研究团队将对简单网络上的潜在问题进行结构分析，评估不同的信息更新机制，并利用这些机制获得见解和生成定制的解决方案。