We are launching our 4 ORAI courses this academic year! 2 in Fall, and 2 in Spring!
Students in the ORAI program will take three out of four graduate courses that teach ORAI and its application to the design of decision-making tools.
Each course is co-developed and co-taught by one faculty in Industrial & Systems Engineering (ISE) with expertise in OR and one faculty in Computer Science (CS) with expertise in AI. Each course integrates technical material with ethics material and involves projects with use cases provided by various USC centers. Students with different backgrounds will work together to succesfully tackle these projects.
Of course you may take any of these interdisciplinary courses even if you are not part of the ORAI program!
Course co-creators and instructors. Phebe Vayanos (ISE) and Evi Micha (CS)
Semester. Fall 2025 (and yearly thereafter)
Catalogue Description. Unified treatment of data-driven decision-making from the lenses of artificial intelligence and operations research for both single- and multi-agent systems.
Course Description. The area of “Data-Driven Decision-Making” (DDDM) is a fast emerging and growing area of scientific research focused on the practice of using data to inform, guide, and enhance decisions in various contexts, ranging from business strategies to public policy, healthcare, marketing, and beyond. The core idea is that decisions that are based on (objective) data enable individuals and organizations to make more informed, evidence-based choices, potentially leading to better outcomes. In this course, we will discuss the transformative role that Artificial Intelligence (AI) and Operations Research (OR) are playing in DDDM in the context of both single-agent systems (where there is only one agent in a defined environment) and multi-agent systems (where decision-making agents interact in a shared environment to achieve common or conflicting goals). The objective of this course is to discuss both the foundations of AI and OR techniques to address decision-making problems integrating data and the latest research, which increasingly integrates techniques from both fields to make effective data-driven decisions. While tools from OR and AI are usually taught in isolation, this course proposes a unified treatment which is valuable to contrast, compare, and train students in identifying good approaches for given situations and in identifying open problems. Through this course, students will develop an appreciation of the value of integrating techniques from the two fields to address decision-making problems informed by data.
The course is intended for Ph.D. students and advanced M.S. students who are ready for the course. It is a required component of the ORAI certificate program.
Recommended Preparation. Optimization (at the level of CSCI 675 or ISE 630); machine learning and its coding in Python (at the level of CSCI 467, CSCI 567, or ISE 529); basic probability (at the level of ISE 220) and linear algebra (at the level of MATH 225 or MATH 235).
Syllabus. [docx]
Course co-creators and instructors. Andrés Gómez (ISE) and Bistra Dilkina (CS)
Semester. Spring 2026 (and yearly thereafter)
Catalogue Description. This course discusses recent advances of using MIO to solve problems in AI/ML, and of leveraging AI/ML to improve MIO solvers.
Course Description. The course will cover recent results on combining the strengths of Machine Learning (ML)/ Artificial Intelligence (AI) and Mixed-Integer Optimization (MIO) technologies. It will discuss how MIO can be used to enhance AI methods in settings with scarce and unreliable data, in high-stakes situations where interpretability and fairness play fundamental roles, and when tackling engineering problems requiring human-AI collaborations. The course will also discuss how AI can be used to improve solving MIO problems, using it to learn ML-based alternatives to key heuristic components of MIO solvers such as branching, understand the key structural properties of a given instance, learn effective configurations of MIO solvers and predict the performance of a given algorithm, thus improving decision-making throughout the solution process.
Recommended Preparation.
An advanced optimization course such as ‘ISE 630: Foundations of Optimization’, ‘ISE 631: Linear Programming’, ‘CSCI 675: Convex and Combinatorial Optimization’, or equivalent
An ML course such as ‘CSCI567: ML’ or equivalent
Knowledge of branch-and-bound algorithms for mixed-integer optimization is recommended, but not required
Coding experience is required [Python will be used in the class]
Syllabus. [docx]
Course co-creators and instructors. Bistra Dilkina (CS) and Vishal Gupta (DSO/ISE)
Semester. Spring 2026 (and yearly thereafter)
Catalogue Description. Models and algorithms for end-to-end learning from both operations research and machine learning perspectives. Applications spanning energy, healthcare, and supply chain. Computational considerations and performance guarantees.
Course Description. Philosophically, ML approaches decision-making under uncertainty by making very few structural assumptions and, instead, learning directly from data. It excels when data are abundant. Conversely, OR zeroes in on specific decision contexts, building models to capture relevant structures and using methods that exploit that structure. OR shines when data are limited, or the problem has complex constraints.
Decision-aware learning is an emerging paradigm in AI that seeks to blend the strengths of both approaches. The best way to do so is still an active area of research. Researchers have proposed using “optimization layers” in deep learning architectures, designing custom convex surrogates for the decision loss of certain classes of optimization problems, debiasing techniques for model agnostic policy learning, and even “light touch” adaptations of traditional predict-then-optimize pipelines. These methods have shown great empirical success in large-scale supply-chain problems, robotic control, energy-systems planning, and more.
In this course, we will develop these approaches rigorously. We pay special attention to algorithmic/computational aspects and their statistical properties/performance guarantees. Much of the course focuses on modeling considerations and how to identify applications that are amenable to an “end-to-end learning” perspective. We aim to empower students to apply state-of-the-art methods and conduct novel research in this area.
The course is intended for Ph.D. students and advanced M.S. students who are ready for the course. It is a required component of the ORAI certificate program.
Recommended Preparation. Students should be familiar with optimization at the level of one of
● ISE 630: Foundations of Optimization
● ISE 631: Linear Programming
● CSCI 675: Convex and Combinatorial Optimization
Students should be familiar with machine learning at the level of one of
● CSCI 567 / ISE 568: Machine Learning
● ISE529: Predictive Analytics
● CSCI559: Machine Learning I: Supervised Methods
Students should also be comfortable with probability concepts common in statistical learning (conditional expectations/distributions and concentration inequalities). Students will likely have been exposed to these concepts as part of a different introductory course in machine learning. Students will be expected to code in Python. Coding assistants (like GitHub Copilot) are permitted and encouraged.
Syllabus. [docx]
NOTE: For this course to count towards your ORAI Certificate, please explicitely ask to replace any of the above courses with it.
Course co-creators and instructors. Vatsal Sharan (CS) and Meisam Razaviyayn (ISE)
Semester. Fall 2025 (and yearly thereafter)
Catalogue Description. Optimization for robustness, fairness, and privacy in machine learning; differential privacy; group fairness, min-max, constrained, and differentially private optimization
Course Description. Optimization techniques still lie at the heart of how models are trained and developed. In this course, we will explore modern considerations such as privacy, robustness and fairness, particularly from the standpoint of optimization techniques. We will both discuss recent research work on formalizing these societal requirements, and algorithmic solutions for obtaining them. Optimization-based approaches such as differentially private optimization, minimax and constrained optimization are particularly useful toolboxes for these problems, and will be explored in their context.
Prerequisites.
Students should have basic optimization knowledge (at the level of CSCI 675 or ISE 630 or equivalent courses)
Students should have basic machine learning knowledge (at the level of CSCI567, CSCI467, ISE 529, or equivalent courses)
Students should have experience in doing ML with Python
Students must be familiar with basic probability and linear algebra concepts.
Syllabus. [docx]