The goal of this project is to develop an online-computation-based approach to dynamic decision making in inventory management, focusing on the newsvendor problem. This fundamental problem in operations research is about choosing how many units of a product to order each period when the product has an uncertain demand and becomes obsolete at the end of each period. The existing approaches to the problem attempt to forecast the demand through stochastic models, but are inadequate when the demand is fast-changing or unpredictable, as is often the case with consumer electronics, seasonal products, perishable food, and some vaccines.
This project will develop algorithms for the newsvendor problem based on the online computation and online learning approaches, in which the objective is to guarantee the performance of the algorithm in the worst-case, as opposed to the average-case used in stochastic approaches. Several variations of the newsvendor problem will be investigated: when only the number of units sold in any period is known and the true demands are censored, when the pricing strategy has to be chosen as well, and when promotions too have to be planned.
The challenge in this project will be to design algorithms that not only guarantee their performance theoretically, but also adapt to rapidly evolving demand patterns observed in the real world. All algorithms developed in this project will be tested on real data from a local supermarket on perishable food products, and compared with the existing stochastic approaches.
This interdisciplinary project will apply a traditional computer science approach to a problem in operations research. As the approach is novel in that domain, this exploratory project will lay the groundwork for more comprehensive projects in the future, which could potentially transform the manner in which supply chain planning under extreme uncertainty is conducted. The project will also expand the understanding of online algorithms and their applicability to decision making. It will help evaluate the relative merits, in the real world, of algorithms designed for the worst-case and those designed for the average-case.
This project's success will improve supply chain planning, and thereby reduce food wastage, at the local supermarket. It has the potential to benefit supply chain planning at a broader scale, for many products all over the world, saving millions of dollars.
