UB workshops will train chemists in AI, thanks to half-million-dollar NSF grant

BUFFALO, N.Y. — Increasingly, much chemistry work happens away from the wet bench or the optical table.

Artificial intelligence like machine learning models can help chemists simulate experiments thousands of times before they’re run or help make sense of their results afterward.

Yet chemistry students often don’t receive the training needed to use many of these advanced software tools.

A University at Buffalo summer program is attempting to change that. A $500,000 grant from the National Science Foundation (NSF) will support campus workshops on cutting-edge computational chemistry techniques that can accurately model quantum dynamics and other complex phenomena occurring within materials.

The grant’s principal investigator is Alexey Akimov, PhD, associate professor in the UB Department of Chemistry who led an NSF-supported pilot version of the workshop from 2021 to 2023.

This new iteration will be a 10-day workshop held once a summer from 2026 to 2029 and taught by leading computational chemists from across the globe. The goal is to reach approximately 100 graduate students, postdocs and early-career scientists from across the country.

In addition, the grant will support the development of a “machine learning for chemists” course at UB that will could be integrated into both the graduate and undergraduate chemistry curriculum.

“Using machine learning to better understand concepts like quantum dynamics and excited-state electronic structure calculations is typically outside of the scope of graduate chemistry programs, so we really want to provide training on this growing ecosystem of tools that figure to only be more widely used in the future,” Akimov says. 

Machine learning models sift through vast amounts of chemistry data to predict how different molecules and materials behave. Some of these tools include Libra, an open-source software that Akimov has been developing for nearly a decade to simulate what happens when a material’s molecules absorb light and enter an excited state.

Better understanding these kinds of dynamics — like the interplay between electrons and nuclei — is crucial for designing the advanced materials needed for solar panels and quantum computers. 

“Instead of doing expensive and tedious tests on many materials, we can run large-scale simulations on many materials under many conditions to help determine the parameters of our experiments,” Akimov says. “Oftentimes it’s the other way around though: We utilize our computational techniques to rationalize recent experiments to provide a better mechanistic understanding of the underlying processes from the quantum mechanical grounds.”.

During the workshops, students will attend lectures and hands-on training sessions for various machine learning models and other computational tools. Many of the featured tools were developed by the workshop’s instructors, who will include Akimov and other leading experts at UB and other universities worldwide. Some of the anticipated instructors come from Stony Brook University, NYU, MIT, the University of Vienna and Los Alamos National Laboratory.

The featured tools require computing power greater than the typical workstation. They’ll be hosted by computer servers at UB’s Center for Computational Research (CCR), which also supported the previous workshops.

By the end of the workshop, students will complete a capstone project where they’ll apply their new machine learning skills to a research area of their choice. In previous workshops, students have modeled solar cell materials and interrogated abstract theories and philosophical questions in quantum mechanics.

“I think students really enjoy this time in Buffalo learning a host of tools and techniques developed by experts here and across the world,” Akimov says. “It’s a really immersive experience.”