Essentially, the engineering of any service-oriented industry, such as ride-sharing or shared-economy systems (think Uber, Instacart, Doordash etc), hospitals, call centers, cloud computing services. However, increasingly, one can see the “Uber-ization” of other industries as well. For instance, manufacturing is undergoing a revolution wherein there are many companies that “share” their manufacturing capacity in (near) real-time; see this: https://qz.com/949147/an-uber-model-for-manufacturing-is-ready-to-upend-the-industry/ for instance.

The engineering of service systems typically involves designing policies and system configurations for maximizing customer satisfaction, while maximizing revenues. For instance, in the case of Uber, the key question is how to match riders and drivers across different geographical locations. This is achieved through ride-pricing, which is decided in real-time. The design of the pricing policy, however, is crucially dependent on having a useful mathematical model of the “Uber system”. Typically, service systems are modeled using large-scale stochastic/queueing networks, where the ‘parameters’ of these models (for instance, the rate at which rides are requested or drivers join/leave the Uber system) are assumed to be fixed in time. In most cases, service systems operate within a random environment; again, using Uber as an example, the demand for rides and drivers can be driven by exogeneous factors such as the weather, or events. Thus, the ‘parameters’ of the model should also be viewed as stochastic in nature. In other words, there is a hierarchical nature to these models. 

The engineering of service systems in increasingly “data-driven” in the sense that it is possible to collect data on the performance of the service system (again, using Uber as an example, performance can refer to questions such as how long does it take for drivers and passengers to be matched?) and use this data for engineering the system. This project specifically focuses on the question of how to use this data to ‘calibrate’ or estimate the stochastic network model of service systems, as a prelude to further engineering of these systems. The complexity of stochastic network models in random environments requires the development of new probabilistic machine learning (ML) methods, and I expect there to be some significant new contributions to that literature as well, in particular around combining ML and ‘physics-based’ stochastic models. The primary thrust of the project is methodological, including providing statistical guarantees on the calibrated models.

Please join us in congratulating Professor Honnappa.