ORIE 6751 - Data-Driven Optimization Under Uncertainty: Theory, Methods, and Current Trends

Spring. 3 credits. Letter grades only (no audit).

Prerequisite: familiarity with basic statistics, probability, and optimization, or permission of the instructor. Enrollment limited to: Ph.D. students. Offered in New York City at Cornell Tech.

N. Kallus.

Optimization with random costs and constraints underlies many important decision-making problems in operations, healthcare, policymaking, and beyond. Models for these problems include stochastic, chance-constrained, robust, and distributionally robust optimization. Recent years have seen intense interest in using data to inform such decision-making models - both data on the uncertain variables themselves and on auxiliary observations. The aim of this course is to understand the landscape of recent developments and prepare students to both use these tools and contribute to them in their own research. The course will combine lectures on the relevant fundamental theoretical constructs and tools with presentations of selected recent papers, clustered into themes, including contextual stochastic optimization, data-driven robust and distributionally robust optimization, optimization of counterfactuals from observational data, and sequential decision making.

Outcome 1: Be able to formulate a decision making problem with uncertain variables as an optimization model.

Outcome 2: Learn the theoretical tools that underlie data-driven optimization and be able to apply them to study both finite-sample and asymptotic properties of data-driven optimization methods.

Outcome 3: Understand the landscape of the current literature and be able to draw upon it in one’s research.

Outcome 4: Become prepared to contribute to the modern literature on data-driven optimization.