Nonparametric Sampling-Based Algorithms for Supply Chain Systems

Regents Of The University Of Michigan - Ann Arbor, Ann Arbor MI

Investigators

Abstract

The research objective of this award is to develop a sampling-based algorithmic framework for sequential decision making problems such as those that arise when supply chain systems experience input certainties, e.g. in demand, capacity, lead time, yield, or product lifetime at the beginning of the decision period. The algorithmic framework can simultaneously learn the input uncertainties through observed data and optimize the system-wide objective on the fly. If successful, the algorithms developed will help decision makers better cope with uncertainties in complex supply chains by analyzing and utilizing data in an online fashion. This project will also help broaden participation of underrepresented groups in research and positively impact engineering education. The models and results of this project will be incorporated in the courses taught by the principal investigator. To achieve the goals of this research, the principal investigator will advance methods to jointly learn and optimize stochastic inventory and supply chain systems, based on past observable data in an online fashion. Such decision rules strike a balance between exploring the underlying random processes and exploiting the processes that have been learned. The principal investigator will provide explicit performance guarantees of the performance of the algorithms. The performance measure is regret, the difference between the objective value of adaptive sampling-based algorithms and the clairvoyant optimal policies that are computed with respect to the true underlying probability distributions, if the latter were known. Among the specific applications, the principal investigator will study inventory and supply chain systems with multiple and perishable products, random capacities, and joint inventory and pricing decisions. If successful, this research will advance the theory and practice of stochastic inventory and supply chain systems for data-rich environments, and provide new decision tools for practitioners.