Audrey Bazerghi
Lecturer of Operations Management, PhD candidate
Contact Information
audrey dot bazerghi at kellogg.northwestern.edu
Faculty advisors
About me
I teach Strategic Decisions in Operations (OPNS454), a full-time/evening and weekend MBA elective at Kellogg. As the main instructor in spring 2024, I also created an AI teaching assistant for the course.
I am a fourth year PhD candidate in Operations Management and a Lecturer at the Kellogg School of Management, Northwestern University. I received the B.Eng. degree in Engineering Physics from Polytechnique Montréal (Canada) in 2015 and the SM/MBA degrees from the Massachusetts Institute of Technology (USA) as a Leaders for Global Operations fellow in 2020. From 2015 to 2018, I worked at Oliver Wyman as a management consultant advising clients in North America on topics focusing on procurement and logistics. In my spare time, I enjoy hiking, climbing and learning new languages.
Working papers
- Relative Monte Carlo for Reinforcement Learning, with S. Martin and G. J. van Ryzin. (2024) Journal version under preparation. Available at: http://dx.doi.org/10.2139/ssrn.4857358
* Accepted at MSOM annual conference 2024, Minneapolis, MN - Last Time Buys during Product Rollovers: Manufacturer and Supplier Equilibria. Bazerghi, A. and Van Mieghem, J. A. (2024). Production and Operations Management 33(3): 757–774. Available at: https://doi.org/10.1177/10591478241231859
* Winner of 2023 POMS College of Supply Chain Management Best Student Paper Competition
* Accepted at MSOM annual conference 2023, Montreal, QC - Last Time Buy or Last Resort? Insights from the Field. Bazerghi, A. and Van Mieghem, J. A. (2022). Available at: http://dx.doi.org/10.2139/ssrn.4114913
Research Interests
My research interests include supply chain management (procurement, inventory control, product life cycle) and optimization & learning (sequential optimization, reinforcement learning).
Current work
I am working on developing a new, general-purpose policy gradient algorithm for reinforcement learning with discrete actions. The algorithm is called relative Monte Carlo (rMC). The policy is improved in real time using relative returns between a root sample path and counterfactual simulated paths instantiated by taking a different action from the root. The method is guaranteed to converge for episodic and average reward tasks. We are testing rMC with a policy network in a two-tiered inventory fulfillment problem and it learns better policies faster than comparable policy gradient algorithms. I will present this work at MSOM 2024 and ISMP in the coming months.
Recent work
I published a research paper with Jan A. Van Mieghem entitled “Last Time Buys during Product Rollovers: Manufacturer & Supplier Equilibria” and a white paper companion “Last Time Buy or Last Resort? Insights from the Field.” When a supplier decides to obsolete a legacy component to focus resources on new growth products, original equipment manufacturers are faced with a difficult situation: how can they continue to support old products with important user bases even after a supplier obsoletes a critical component? Last time buys are one of the most widespread solutions. Our research paper uses game theory to study how supplier and manufacturer interact during a negotiation that considers both the last time buy of an old part and the launch of its successor.