Cornell researchers are demonstrating how artificial intelligence - particularly deep learning and generative modeling - can accelerate the design of new molecules and materials, and even function as an autonomous research assistant.
In a study published April 9 in Advanced Science, researchers explored how to make AI models more efficient and effective in predicting the properties of molecules for everything from drug development to materials design. The team focused on a technique called knowledge distillation, which involves compressing large and complex neural networks into smaller, faster models.
The distilled models ran faster - and in some cases improved performance - while working well across different experimental datasets, making them ideal for molecular screening without the heavy computational power required by most AI systems.
"To accelerate discovery in materials science, we need AI systems that are not just powerful, but scientifically grounded," said Fengqi You, the Roxanne E. and Michael J. Zak Professor in Energy Systems Engineering in Cornell Engineering, who co-authored the study with graduate student Rahul Sheshanarayana. "Our work shows that AI can learn to reason across chemical and structural domains, generate realistic materials, and model molecular behaviors with efficiency and precision - all while aligning closely with the fundamental principles of materials science."
You directs the Cornell AI for Sustainability Initiative and co-directs the Cornell University AI for Science Institute - two programs advancing the next generation of AI-powered science. Both have supported forward-thinking efforts in You's research group.
In a Nature Computational Science paper published May 9, You and Zhilong Wang, an Eric and Wendy Schmidt AI in Science Postdoctoral Fellow, introduce a new framework for generative inverse design of crystalline materials. Crystals, with their repeating atomic patterns and strict symmetry, present a challenge for AI models, which often rely on abstract or oversimplified representations.
The research group's proposed solution: a physics-informed generative AI model, which embeds crystallographic symmetry, periodicity, invertibility and permutation invariance directly into the model's learning process. The framework allows AI to generate novel crystal structures that are not only mathematically possible, but chemically realistic.
"Our goal is to ensure that AI-generated materials are scientifically meaningful," Wang said. "We're encoding physical principles and operating conditions directly into the learning framework, so instead of relying on massive trial-and-error, we're guiding the AI with domain knowledge."
In a review paper published May 12 in Advanced Materials, You and doctoral student Wenhao Yuan detail an emerging class of AI systems known as generalist materials intelligence. Unlike traditional models trained for specific tasks, generalist materials intelligence is powered by large language models and interacts with both computational and experimental data to reason, plan and interact with scientific text, figures and equations, functioning as an autonomous research agent.
"What's exciting is the idea that AI can start to engage with science more holistically," Yuan said. "We're teaching AI how to think like a scientist, developing hypotheses, designing materials and verifying results."
The studies were supported in part by the Eric and Wendy Schmidt AI in Science Postdoctoral Fellowship, a program of Schmidt Sciences.
In parallel with the group's research, You is also bringing AI concepts into the classroom. This spring, he launched a new graduate-level course, AI for Materials, which introduces students to new techniques for materials science, including deep-learning applications in energy storage, synthesis optimization and materials behavior modeling.
"The course emphasizes transformative applications and the challenges of applying AI to accelerate materials design," You said. "It's about equipping the next generation of researchers and engineers with the knowledge to drive innovation at the intersection of AI and materials science."
Syl Kacapyr is associate director of marketing and communications for Cornell Engineering.
