RESILIENT NETWORKS
POMS International Conference 2022 Budapest

Big data provides new opportunities to tackle one of the main difficulties in decision-making systems – uncertain behavior following an unknown probability distribution. Standard data-driven approaches usually consist of two steps. The first step involves predicting or estimating the uncertainty behavior using data. Then the second step requires finding decisions that optimize an objective function that depends on the output of the first step. Instead of the classical two-step predict-then-optimize (PTO) procedure, this tutorial examines data-driven solutions that integrate these two steps. We first introduce the problem formulation as a contextual stochastic optimization. In this formulation, the objective function depends on the unknown uncertainty and the distribution of the uncertainty is associated with some contextual information. Massive data is often available to solve this problem, including historical observations of the uncertainty and contextual information. Therefore, machine learning tools have become an important technique to achieve integrated data-driven solutions. Yet, it is noteworthy that the goal of the integrated data-driven solution is very different from traditional predictive tasks for machine learning. Moreover, different integrated data-driven methods have shown applicability and effectiveness in many real-world decision-making situations, including inventory management, COVID-19 pandemic, and power system. To demonstrate the practicality and the real-world impact, we review current achievements of integrated methods in different real-world applications in operations management.