ICME Research Symposium 2026: AI for Science and Engineering
Not Just Faster Science: How AI Is Beginning to Change Discovery
AI systems that read scientific literature, propose hypotheses, design experiments, and identify hidden mathematical structures are beginning to influence how research is conducted across science and engineering.
On May 19, 2026, ICME hosted its annual Research Symposium on the theme of AI for Science and Engineering, bringing together researchers, industry leaders, national laboratory scientists, faculty, and students to examine AI's growing role in scientific and engineering research. For decades, scientific computing has transformed research by making simulation and analysis dramatically faster and larger in scale. This year's symposium focused on a broader shift: AI is reshaping how questions are asked, how experiments are designed, and how results are interpreted across disciplines. The audience remained engaged throughout the day, with discussions extending well beyond the scheduled close.
Stanford is home to a broad range of leading work in artificial intelligence, and the ICME symposium highlighted a distinctive focus within that landscape: the mathematics, algorithms, high-performance computing, and hardware systems that underpin scientific and engineering discovery. Few forums bring together theory, computation, hardware, and domain science across areas ranging from genomics to cosmology within a single conversation.
ICME at the Intersection of AI, Computation, and Discovery
The symposium opened with remarks from Eric Darve, Professor of Mechanical Engineering and Director of ICME, who described AI for science as one of the defining interdisciplinary challenges in modern research. He emphasized the importance of developing models that are scientifically interpretable, reliable under uncertainty, and grounded in physical laws.
Darve outlined ICME’s position across mathematics, computation, machine learning, and domain science, arguing that future scientific advances will depend on integrating these areas with advanced computing platforms and large-scale scientific datasets. He also emphasized ICME’s interdisciplinary structure across Stanford and its role in training students at that intersection while connecting academia, industry, and national laboratories through research and education initiatives.
Highlighting graduate student research spanning healthcare, quantum computing, finance, generative modeling, and GPU-aware numerical algorithms, Darve noted that many of the projects shared common themes, including mathematical structure, interpretability, scalability, physics-informed AI, and hardware-aware computation, reflecting core areas of ICME research and training.
Is AI What’s Next for Engineering?
Gianluca Iaccarino, Robert Bosch Chair and Professor of Mechanical Engineering, opened the morning keynote with a question that framed his entire talk: Is AI What's Next for Engineering?
Iaccarino discussed how AI is reshaping simulation, design, and engineering workflows while emphasizing that intuition, judgment, and physical understanding remain central to the practice. Comparing current AI systems to early-career researchers, he noted that while these systems can already contribute meaningfully to research, they still require guidance, interpretation, and domain expertise from scientists and engineers.
He also drew on the history of science to examine how shifts in understanding have repeatedly reshaped broader ideas about knowledge and discovery.
From Wafer-Scale Computing to Scientific Discovery
Andrew Feldman, Co-Founder and CEO of Cerebras Systems, delivered the industry keynote just days after his company’s IPO, reflecting how seriously the AI industry now views the infrastructure question. Feldman traced the original vision behind Cerebras: that modern AI and scientific computing workloads required a fundamentally different architecture from traditional GPU-based systems, a conviction that shaped the development of the company's wafer-scale engine. He also reflected on his time at Stanford and the early stages of building Cerebras Systems, encouraging students to pursue the problems most people assume are impossible.
A fireside discussion with Iaccarino, Feldman, and Darve expanded on many of these themes.
Vision Talks Highlight Emerging Directions in AI and Scientific Computing
Faculty vision talks throughout the day highlighted research directions across theory, computation, and data-driven discovery.
In the morning session, Surya Ganguli presented theoretical work on how generative AI systems learn language and visual structure, including findings on how model accuracy scales with data and the development of universal scaling laws. Vasilis Syrgkanis introduced a framework for benchmarking how AI systems reason about causal identification, research design, and estimation in real-world empirical settings.
The afternoon session ranged from generative modeling and genomics to astrophysics: Renyuan Xu on one-step generative modeling via gradient flows, Anshul Kundaje on decoding regulatory DNA syntax and interpreting genetic variation with deep learning, and Risa Wechsler on how AI and simulation-based inference applied to surveys like the Rubin Observatory's LSST are turning massive datasets into new discoveries about dark matter, galaxy evolution, and the structure of the universe.
PhD Spotlights gave graduate students the opportunity to present research on scientific reasoning with vision-language models, score-debiased kernel density estimation, and multiscale modeling of sediment-controlled ice-sheet motion.
The symposium also recognized outstanding student posters, with attendees moving through a broad range of projects spanning computational mathematics, scientific machine learning, optimization, climate science, biomedical applications, physics-informed modeling, and large-scale simulation.
From the Genesis Mission to AI-Assisted Physics Research
Rajeev Thakur, Deputy Division Director and Argonne Distinguished Fellow at Argonne National Laboratory, brought a national perspective to the afternoon, presenting on the Department of Energy's Genesis Mission, an initiative focused on infrastructure for AI-driven discovery, energy innovation, and national security applications.
Alex Lupsasca, Research Scientist at OpenAI, closed the keynote program with a talk on accelerating scientific discovery through AI. Drawing on his work in theoretical physics and black hole research, Lupsasca described how frontier models can compress complex calculations and surface structure that would take researchers years to find by hand. He also reflected on efforts to reproduce and extend results involving hidden symmetries in black hole physics as an example of human-AI collaboration in scientific research.
Building Connections
Between sessions, the symposium created opportunities for networking and conversation. The networking lunch and closing reception brought together students, faculty, industry engineers and national laboratory researchers who do not often share the same space. The poster session also gave graduate students an opportunity to present their work directly to that audience, with awards presented to the top three student posters. For many attendees, these informal moments proved just as valuable as the formal program.
