AI for Science course bridges disciplines

A new course at U-M teaches students to integrate AI with scientific methods across various fields.
Alex Rodriguez teaches a class. He holds an ipad in his hands.
Prof. Alexander Rodríguez’s AI for Science course teaches students how to apply the latest machine learning methods in various scientific disciplines.

As artificial intelligence (AI) technology advances at a breakneck pace, a pioneering course at the University of Michigan is equipping students to harness its potential within a scientific context. Launched in Fall 2024, CSE 598: AI for Science marries the latest AI technology with scientific methods, empowering students to design and apply state-of-the-art AI tools across a range of disciplines.

Designed and taught by Alexander Rodríguez, assistant professor of computer science and engineering at U-M, AI for Science integrates AI techniques with scientific priors, including equations, simulations, causal relations, and more, providing students with a comprehensive view of recent methodological advances and challenges across the scientific world. 

“While many courses teach AI fundamentals, few address its recent developments in tackling the technical challenges found in scientific and engineering domains,” said Rodríguez. “What makes this course innovative is its interdisciplinary approach—it gives students the tools to tackle complex scientific problems with AI, while remaining grounded in core scientific principles.”

A unique approach

AI for Science provides students with a deep dive into a variety of advanced topics related to AI, beginning with a review of deep learning principles before progressing to specialized subjects such as physics-informed neural networks, neural operators, differentiable simulators, and knowledge-guided learning algorithms. The course also highlights techniques for learning scientific equations and discusses the role of foundation models in scientific applications. Together, this broad but detailed curriculum gives students both theoretical and practical insights into the fundamentals of AI and its application in scientific exploration.

Alex Rodriguez stands at the front of a lecture hall, delivering a lecture. Behind him is a powerpoint slide projected onto a screen with the heading "AI-guided generation of scientific hypotheses"
The methods taught in AI for Science can be applied across scientific and engineering domains.

Part of what sets the AI for Science course apart is its highly interdisciplinary approach. The course attracts graduate and advanced undergraduate students from diverse fields, including not just computer science, but also physics, biomedical engineering, aerospace engineering, and climate sciences. This diversity of academic backgrounds enables students to learn from each other’s expertise and address problems from different perspectives.

“One strength of this course is that students have the ability to interact with people from various disciplines,” said Rodríguez, “These interactions provide a richer educational experience that reflects the collaborative nature of modern scientific research.”

In addition, the course features visiting lecturers from leading institutions such as MIT and Virginia Tech, who present recent developments and applications in AI. Students also engage with the latest research by reading and discussing new papers on innovative methodologies and technologies. These expert visits, along with exposure to current research, provide students with unique insights into the field and ensure that the curriculum remains up-to-date with the latest advancements in AI.

Innovative projects with real-world applications

Students in AI for Science are encouraged to bring their research interests into the classroom, culminating in final projects that reflect the wide applicability of AI methods. Projects in the Fall 2024 course ranged from computational biology and epidemiology to mechanical engineering and climate modeling.

For instance, Mukundh Murthy, a senior in computer science, applied physics-informed neural networks to study cell signal transduction. His project focused on modeling the dynamics of how cells respond to a given signal within varying timespans, a process essential in understanding why similar cells in a group can respond differently to the same treatment. By incorporating known equations for different classes of biological signaling into machine learning models, Mukundh’s work aimed to predict transduction timelines more accurately and contribute to more interpretable and effective biological models of signaling.

“Machine‑learning tools in biology often feel like opaque ‘black boxes’—you get an answer without understanding the detailed steps of the biological processes involved,” said Mukundh. “On the other hand, this course helped me explore the applications of physics-informed machine learning to better understand fundamental biology.”

Equipping students with the skills to integrate AI into scientific domains is crucial in preparing them for their careers.

Alexander Rodríguez, Assistant Professor of Computer science and Engineering

CSE PhD student Álvaro Vega Hidalgo explored climate prediction models in his project, leveraging physics-informed AI to improve the accuracy and explainability of these models. His research integrated physical laws, such as atmospheric dynamics, into AI frameworks to enhance the reliability of short- and long-term climate forecasts. This innovative approach not only improved prediction accuracy but also provided insights into the behavior of complex climate systems.

“By integrating physical laws into AI, we can make more reliable climate predictions, which are crucial in addressing climate change impacts,” Alvaro noted.

These projects underscore the course’s emphasis on domain-agnostic AI, demonstrating how AI techniques can be adapted to solve problems across a wide array of scientific fields, from understanding cellular processes to predicting environmental changes.

A hands-on learning experience

Going forward, Rodríguez aims to deepen students’ educational experience through additional interactive course elements, such as additional guest lectures and role-playing sessions, where students will propose and critique AI methods from various perspectives. Rodríguez recently secured a U-M Enhancing Engineering Education (E3) grant to further develop these interactive activities.

“Role-playing activities allow students to take on different roles and viewpoints, helping them evaluate AI methods from both a computational and scientific perspective,” Rodríguez explained. “This helps prepare them for the collaborative, interdisciplinary settings they’ll encounter in their careers.”

AI for science and engineering

CSE 598: AI for Science exemplifies the burgeoning movement to integrate AI into science and engineering curricula at U-M. The course’s success in doing so is reflected in its high enrollment from various disciplines and positive student evaluations, with several students praising the course’s interdisciplinary approach, interactive projects, and guest lectures.

As AI continues to redefine scientific research, courses like AI for Science play a crucial role in preparing the next generation of innovators. “Equipping students with the skills to integrate AI into scientific domains is crucial in preparing them for their careers and driving future breakthroughs across disciplines,” said Rodríguez.