Exploring AI Frontiers in Science and Engineering
The Stanford Institute for Computational & Mathematical Engineering hosted its annual research symposium on May 20, 2025, bringing together participants from academia and industry to engage in the current developments in computational mathematics, machine learning and data science.
The event opened with remarks from Professor Eric Darve, Faculty Director of ICME, who outlined the institute’s mission to connect theoretical foundations with practical applications. He discussed ICME’s origins and its evolution into a center for interdisciplinary research—advancing work in climate modeling, financial systems, personalized medicine, and energy systems. Professor Darve emphasized the institute’s role in bridging theory and application, stating, “We are not just using AI, but understanding the mathematical foundations that will power tomorrow’s breakthroughs.” He highlighted how ICME unites physical insight with data-driven discovery.
The morning session focused on physics-based machine learning approaches that integrate scientific principles with data-driven methods. Professor Charbel Farhat delivered the morning academic keynote, “Physics-Based Machine Learning: What, Why, How, and Impact,” where he analyzed applications across a range of systems, including aircraft structural analysis, parachutes for Mars landing, and Formula 1 cars. He described how first-principle physics-based methods can be seamlessly combined with data-driven models to achieve better predictions and understanding.
Professor Lexing Ying’s afternoon keynote, “Computational Mathematics Meets Quantum Algorithms,” explored how advanced numerical methods can enhance quantum algorithms. He demonstrated how classical tools such as the Fourier transform, Prony’s method, and the Weiss algorithm can be adapted to solve critical tasks like quantum phase estimation and quantum signal processing. His work highlighted that computational mathematics combined with quantum algorithms can lead to faster, more accurate simulations, pushing the boundaries of what’s possible in quantum computing.
As the industry keynote speaker, NVIDIA Chief Scientist Bill Dally emphasized how advances in hardware, particularly GPU architectures, are driving the rapid progress of AI in science. He highlighted the critical role of scalable, high-performance computing hardware in enabling researchers to train larger models, run faster simulations, and unlock new possibilities in data-driven discovery.
Faculty vision talks throughout the day highlighted the diverse ways ICME researchers are tackling today’s most pressing scientific challenges. In the morning session, Daniel Tartakovsky presented novel methods for assimilating binary-sensor data, while Aditi Sheshadri shared high-resolution simulations of atmospheric gravity waves to better understand climate dynamics. James Zou shared how AI agents could play a more active role in the scientific process itself, from research design to discovery. In the afternoon, the focus shifted toward data-driven decision-making and interpretability. Markus Pelger examined the impact of AI-driven trading on market stability, while Jose Blanchet offered strategies for making optimal decisions when models are uncertain or imperfect. Ramesh Johari presented new findings on experimentation in digital marketplaces, and Lei Xing introduced interpretable deep learning techniques for tabular and graph-based data. These talks reflected ICME’s core strength in bridging theoretical foundations with real-world applications across disciplines.
The symposium featured PhD lightning talks and a poster session where ICME MS and PhD students presented research spanning causal representation learning in language models to approaches for equitable congestion pricing systems.
A big thank you to all who participated. It was a stimulating and engaging event! Photos from the event can be viewed here.
