Two Problems in Resource Management: Scheduling with Prediction Errors and Fair Data-Driven Allocation with Limited Data

Abstract

Resource management is challenging when one needs to allocate scarce or limited resources to different entities with heterogeneous demands. In many practical situations, predictions of relevant quantities are only possible or available. While accurate estimates can certainly allow for better decisions to be made, a key challenge is to extract the maximum benefit when highly accurate estimates are not available (or possible). Even in the presence of reasonable estimates, other factors such as the need for a timely or real-time resource allocation can add to the complexity of the resource management process. This thesis studies two problems in resource management: scheduling with prediction errors and fair data-driven allocation with limited data. These problems both consider scenarios where only estimates of the demand are known and real-time resource allocation is required. In the second problem, the supply for resources is also not known within the decision-making period and is estimated.