Grants to spur innovations in generative AI, computational science


Research teams at the University of Michigan will share $575,000 to facilitate high-impact research across a broad range of domains, including sustainable energy, pandemic response, ultra-powerful computing, and generative artificial intelligence for science and education.

The Michigan Institute for Computational Discovery and Engineering, a unit within the Office of the Vice President for Research, recently awarded funds to seven research teams as a part of its Catalyst Grants program.

“This year’s Catalyst Grants focus on the development of novel generative artificial intelligence techniques for scientific applications. These projects are not just highly innovative, but also strategically important to U-M, particularly with respect to our evolving relationship with national laboratories,” said MICDE Director Karthik Duraisamy, professor of aerospace engineering and of mechanical engineering.

“U-M is leading the narrative in the development and deployment of generative AI across the spectrum of academic missions, and these projects are directly responsive to that ideal.” 

Since its Catalyst Grants program was launched in 2017, MICDE has supported a broad spectrum of research in computational science and engineering.

“Our objective is to initiate these projects and support U-M researchers from the development of their conceptual ideas to making a global impact,” said Vancho Kocevski, managing director of MICDE.

MICDE awarded grants for the following projects for 2023-24:

Efficient diffusion models for scientific machine learning 

Liyue Shen, assistant professor of electrical engineering and computer science; Jeff Fessler, William L. Root Collegiate Professor of Electrical Engineering and Computer Science and professor of applied physics, of biomedical engineering and of radiology; and Qing Qu, assistant professor of electrical engineering and computer science, will develop novel techniques to improve the training and sampling efficiency of generic diffusion models and introduce computationally efficient diffusion models for high-dimensional data to further enhance data, memory and time efficiency.

Simulating the impact of early outbreak uncertainty on pandemic response intervention policies

Michael Hayashi, clinical assistant professor of epidemiology; and Joseph N.S. Eisenberg, professor of epidemiology and of global public health, will develop a novel interdisciplinary approach that draws from modeling frameworks from both health and social sciences to capture the feedback between policymakers, members of the public, and disease transmission to prepare policymakers for the next pandemic.

Computational modeling of a coupled aero-hydro-structural-mooring integrated dynamic system with deep learning for floating offshore wind turbine design

Jeremy Bricker, associate professor of civil and environmental engineering; Ayumi Fujisaki-Manome, associate research scientist at the Cooperative Institute for Great Lakes Research; Seymour Spence, associate professor of civil and environmental engineering; and Yulin Pan, assistant professor of naval architecture and marine engineering, will develop a novel high-fidelity and computationally efficient approach to predict the dynamic response of floating offshore wind turbines and a short-term early warning technology in extreme sea conditions based on deep learning.

Program synthesis tools for scientific computing

Brendan Kochunas, assistant professor of nuclear engineering and radiological science; and Xinyu Wang, assistant professor of electrical engineering and computer science, will use program synthesis methods to take mathematical algorithm descriptions as input and produce functionally correct and performant code using machine learning, a novel approach in scientific computing.

An AI coachbot for strengthening self-regulated learning of computational ML and AI

Raj Rao Nadakuditi, associate professor of electrical engineering and computer science, will develop a generative AI-based feedback and coaching chatbot that can transform the training of hundreds of computational scientists- and engineers-in-training at U-M.

Foundation models at quantum mechanical accuracy for battery electrolyte design using wafer-scale computing

Venkat Viswanathan, associate professor of aerospace engineering; and Vikram Gavini, professor of materials science and engineering and mechanical engineering, will train a foundation model on the largest available chemical dataset to achieve accuracy similar to quantum mechanical computational methods and will fine-tune the model for tasks relevant to electrolyte design.

Resilient distributed training of large models

Mosharaf Chowdhury, associate professor of electrical engineering and computer science, aims to develop Oobleck, a resilient hybrid-parallel training framework, to enable resilient distributed training of large generative AI models with consistently high throughput even in the presence of failures.


Leave a comment

Commenting is closed for this article. Please read our comment guidelines for more information.