Feizi Receives Prestigious ARO Early Career Program Award

Soheil Feizi, an associate professor of computer science at the University of Maryland, has been awarded $360,000 by the Army Research Office (ARO) to develop provable methods that can improve the robustness of dynamic AI systems.

The funding comes from ARO’s prestigious Early Career Program Award, which incentivizes early career university faculty to pursue fundamental research.

Feizi’s proposal, “Provably Robust Dynamic Systems,” was funded for three years. Its goal is to provide a comprehensive and fundamental understanding of provable robustness in dynamic and adaptive learning setups, leading to practically useful methods with theoretical guarantees.

Robotics and autonomous planning—critical in supply chain management scenarios—utilize these types of dynamic systems with deep neural networks, and its applications are increasing, Feizi says.

However, these dynamic models can be very sensitive to small adversarial perturbations, he explains, where adversaries can induce time-dependent alterations at multiple scales to change the outputs of the systems.

These adversarial perturbations can be small and imperceptible to humans, making the security risks even more worrisome in highly sensitive applications where model reliability and trustworthiness are critical.

Most of the existing research in this area is on the robustness of static models, such as image classification problems, Feizi says. But these techniques fall short against strong and adaptive adversarial attacks that are particularly designed to break dynamic learning systems. This is because in dynamic settings, the adversary can adapt its strategy to the defense applied by the victim model in previous time steps.

Feizi expects his innovative methods will address this issue, leading to AI systems that are able to function effectively even when operating in unexpected or changing environments.

The ARO funding is the latest early career award for Feizi, who has an appointment in the University of Maryland Institute for Advanced Computer Studies and is a core faculty member of the University of Maryland Center for Machine Learning.

In 2022, he received an Office of Naval Research Young Investigator Award to advance artificial intelligence agents that can seamlessly and robustly interact with humans and perform various tasks with minimal supervision.

And in 2020, he received a National Science Foundation Faculty Early Career Development (CAREER) award for a project intended to study deep generative models.

—Story by UMIACS communications group