Newswise – Two Georgia Tech PhD students have created a fully accredited student-run faculty alumni course linking mathematics, engineering and machine learning.
With the support of Michael Clamkin, a student researcher at the National Science Foundation AI Institute, Andrew Rozanberg designed the course to fill the gaps he saw in existing classrooms.
“While Georgia Tech offers excellent courses in optimization, control and learning, we have not found a single class that connects all of these disciplines in a cohesive way,” Rozanberg said. “In our study, it was clear that these topics were deeply interrelated.”
Problem-driven learning
The course starts with the underlying problems and goes against the methods needed to solve them. Rosemberg said the approach was intentional. He said that courses often focus on the solitary method rather than showing how methods contribute to a larger context. This will keep the course focused on problem-driven discovery.
This class also serves as a way for Rosemberg and Klamkin to enhance their own teaching and teaching skills.
Goals and structure
The main goal of this course is to ensure students have a clear understanding of how machine learning techniques, such as mathematical programming, classical optimal control, and reinforcement learning, connect with each other. Students are also working to produce structured books by the end of the semester.
“The hope is that this resource not only solidifies our own learning, but also serves as a guide for other students who want to approach these issues in the future,” Rozanberg said.
Responsibility is distributed among participants, with each student giving lectures, reviewing the work of his peers, and contributing to collective discussion. Rosemberg and Klamkin will provide additional support when needed, but AI4opt faculty and director Pascal Van Hentenryck ensures that the class remains aligned with the broader academic goals.
Student Ownership and Collaboration
Rosemberg noted that student-driven models give students a deeper sense of ownership, take responsibility for their own learning and have a stronger impact. This model allows students to determine what they learn and why. This encourages critical thinking.
This course uses GitHub as its primary workflow platform. Rosemberg said it will add transparency and prepare students for real-world research practices.
“Github works in the same way as university systems like Canvas and Piazza, and it also has the advantage of making all your contributions visible to the world,” Rosemberg explained. “This will help students gain pride and ownership of their work and introduce Git, an essential tool for software development and modern STEM research.”
New insights and challenges
Students began coordinating course themes and research, including forming eligible exam topics within operational research intersections, optimal control, and supplementary learning. Rosemberg said exploring the comparative strengths of these areas side by side is one of the most rewarding results.
Balancing independence and guidance has proven to be the biggest challenge. He said they have learned to evolve in real time with students and emphasize mutual responsibility to promote collective progress in the class.
Looking ahead
Rosemberg said future iterations of the course could be more focused on setting expectations early given the efforts needed to lecture in this format.
His advice to anyone who wants to replicate a model is to focus on building a dedicated core team.
“We start with a small, motivated group,” Rozanberg said. “Like a startup, success depends on structure and on the dedication of the stakeholders.”
