Skip to main content

Associate Professor of Computer Science Marco Levorato was awarded $500,000 from the National Science Foundation (NSF) for his grant, “A Design Automation Methodology for Flexible Real-Time Computing based on Split and Early Exit Neural Models.” Levorato is partnering with Associate Professor of Electrical Engineering and Computer Science Mohammad Al Faruque, making this a three-year grant collaboration between the Donald Bren School of Information and Computer Sciences (ICS) and the Samueli School of Engineering.

“At a very high level,” says Levorato, “the goal of the project is to build automated design solutions for flexible real-time computing in mission critical applications such as autonomous vehicles and mobile healthcare.”

Leveraging split computing (SC) and early-exit computation (EEC), the researchers hope to bridge runtime system optimization with advanced deep learning (DL) model architectures. They will design their DL models to adapt real-time data analysis to a system’s time-varying characteristics, such as computing power and available energy, and to the information stream. This should boost the performance of critical applications ­— including AI-empowered monitoring for mobile health — while reducing energy consumption and wireless channel usage.

— Shani Murray