Smarter data-driven decision-making by integrating prediction and optimization
Big data provides new opportunities to tackle one of the main difficulties in decision-making systems – uncertain behavior driven by the unknown probability distribution. Instead of the classical two-step predict-then-optimize (PTO) procedure, we provide smarter data-driven solutions by integrating these two steps. In the first half of this talk, we focus on a multi-period inventory replenishment problem with uncertain demand and vendor lead time (VLT), with accessibility to a large quantity of historical data. Different from the traditional two-step predict-then-optimize (PTO) solution framework, we propose a one-step end-to-end (E2E) framework that uses deep-learning models to output the suggested replenishment amount directly from input features without any intermediate step. The E2E model is trained to capture the behavior of the optimal dynamic programming solution under historical observations, without any prior assumptions on the distributions of the demand and the VLT. This algorithm is currently implemented in production at JD.com to replenish thousands of products. In the second half of this talk, I will move to a more general setting of the contextual stochastic optimization problem. We propose an integrated conditional estimation-optimization (ICEO) framework that estimates the underlying conditional distribution using data while considering the structure of the downstream optimization problem. We show that our ICEO approach is asymptotically consistent and further provide finite performance guarantees in the form of generalization bounds. We also discuss the computational difficulties of performing the ICEO approach and propose a general methodology by approximating the potential non-differentiable oracle. We also provide a polynomial optimization solution approach in the semi-algebraic case. The concept of E2E, which uses the input information directly for the ultimate goal, shortens the decision process and can also be useful in practice for a wide range of circumstances beyond supply chain management.
Meng Qi is a Ph.D. Candidate in the Department of Industrial Engineering and Operations Research at University of California, Berkeley, where she is advised by Prof. Zuo-Jun (Max) Shen. Previously, she graduated from Tsinghua University with a B.S. in Physics. Her research focuses on developing more automatic and robust data-driven solutions for decision-making with uncertainty, combining tools and concepts from optimization, machine learning, and statistics. From an applications perspective, her research focuses on supply chain management and retail operations. As a part of it, she actively collaborates with industrial partners in e-commerce.