The development of machine learning models that can predict the properties of unseen and yet-to-be-discovered molecules and materials is poised to play a critical role in addressing humanity’s great challenges. NYU Arts & Science will be at the forefront of that effort through the work of Stefano Martiniani and the team he leads, which was recently awarded a $4.5M grant from the National Science Foundation.
“Large-scale machine learning models used on language and visual tasks have seen spectacular achievements,” noted Martiniani. “With its focus on materials and molecular discovery, our project will have a dramatic impact on scientific research and the development of materials to address societal challenges such as clean energy conversion and energy storage, carbon capture, water purification, drug delivery, and advanced materials for quantum and aerospace technology.”
Machine learning (ML) falls within the larger category of artificial intelligence. Using algorithms, machines learn from data without explicit human intervention. Martiniani’s team will employ an emerging computational paradigm known as “foundation models” (examples include ChatGPT and DALL-E) which “trains” the model on extremely large amounts of diverse and easily available data. Vast quantities of data require “cyber” infrastructure, and for that the team will leverage and expand the ColabFit Exchange, the database they developed with a previous NSF grant. ColabFit Exchange is the first and largest public database of its kind, specializing in data for machine learning interatomic potentials.
In a measure of the grant’s significance, U.S. Senate Majority Leader Chuck Schumer and U.S. Senator Kirsten Gillibrand hailed the award in a joint press release. “I’m proud to support this federal funding for New York University, one of our country’s leading research institutions, to develop machine learning models,” said Senator Schumer. “NYU’s project will ensure that New York, and the country, remain at the cutting edge of developing new technologies, like machine learning, to benefit scientific research, educational advancement and economic development.”
As Principle Investigator, Martiniani leads a team of scientists representing three universities and an industry partner. The team comprises co-PIs Ellad Tadmor (Dept. of Aerospace Engineering & Mechanics, University of Minnesota), George Karypis (Dept. of Computer Science, University of Minnesota), Adrian Roitberg (Dept. of Chemistry, University of Florida), Mingjie Liu (Dept. of Chemistry, University of Florida), Richard Hennig (Dept. of Materials Science & Engineering, University of Florida), Mark Transtrum (Dept. of Physics, Brigham Young University), and Huzefa Rangwala (Amazon Web Services).
Stefano Martiniani received his B.Sc. from Imperial College London in 2012, followed by an M.Phil. in Scientific Computing (2013) and Ph.D. in Chemistry from the University of Cambridge in 2017. He spent two years as a postdoctoral associate in the Center for Soft Matter Research, Department of Physics at NYU Arts & Science, and subsequently joined the University of Minnesota as an Assistant Professor of Chemical Engineering and Materials Science (2019-2021). In 2022 he returned to NYU Arts & Science as Assistant Professor of Physics, Chemistry, and Mathematics.
Martiniani is the recipient of several awards, including the Simons Foundation Faculty Fellowship, Gates Cambridge scholarship and an Outstanding Thesis Prize from the University of Cambridge. This month, the International Union of Pure and Applied Physics awarded Martiniani the Interdisciplinary Early Career Scientist Prize, for “groundbreaking contributions to the understanding of the statistical mechanics of active and amorphous systems via the development of uniquely original approaches for quantifying order, entropy and entropy production in systems far from equilibrium, including granular and active matter, neural networks and biological systems.”