Three A&T Researchers Aim to Solve AI Reliability with $493K DOD-NSA Award

By Alexander Saunders / 09/24/2021 Research and Economic Development, College of Engineering

EAST GREENSBORO, N.C. (Sept. 24, 2021) – Three faculty researchers in the College of Engineering (COE) at North Carolina Agricultural and Technical State University are addressing issues of national security regarding the integration of artificial intelligence (AI) solutions into critical domains – including autonomous vehicles, robots, defense systems, critical battlefield infrastructure monitoring, complex traffic flow management and more. The research addresses the risk of adversarial attacks in the complex network of data in society’s increasing number of modern smart devices, automobiles and financial systems.

With a two-year, $493,957 grant from the National Centers of Academic Excellence in Cybersecurity (NCAE-C), sponsored by the Department of Defense’s National Security Agency (NSA), the team of researchers representing the Department of Computer Science (CS) and the Department of Computational Data Science and Engineering (CDSE) will research effective ways to modernize AI systems and networks that are vulnerable to unexpected behaviors and adversarial attacks.

The grant includes a year three option for additional support upon successful completion of the year two initial goals.

The award comes ahead of the Spring 2022 opening of the $90 million Harold L. Martin Sr. Engineering Research and Innovation Complex, designed to foster engineering and computer science collaborations that address the nation’s most critical problems.

“To address these critical challenges, we are building secure and trustworthy AI for Cyber Physical Systems (CPS) for the NSA,” said Kaushik Roy, Ph.D., a CS associate professor who serves as the principal investigator on the interdisciplinary project. “Our multidisciplinary team of faculty and students from A&T and Florida A&M University (FAMU) will develop and enhance cybersecurity as well as AI and machine learning techniques for CPS in ways that are relevant to the nation’s defense.”

Cyber physical systems research investigates the fusion of sensing, computation, networking and physical processing controlled by algorithms. The protection provided by cybersecurity innovations is essential for the deeply complex computational relationships of these systems.

This is a photo of lab equipement

Society is flooded with data, said Roy, and even the most modern and robust AI systems and networks are vulnerable to unexpected behaviors and adversarial attacks.

“With this research, we will strengthen the operation of highly automated AI systems to withstand the vulnerability for unexpected or adversarial behaviors,” said Balakrishna Gokaraju, Ph.D., a CDSE associate professor. “This work will enable cybersecurity for national media, political campaigns, newspapers, and social media content.”

Albert Esterline, Ph.D., a CS associate professor, is leading the “trustworthiness” between the AI systems aspects of the project using a “handshake” approach where two systems can verify the other is secure and work together. As data becomes increasingly more available and interconnected, it is critical for practitioners to have their systems recognize data branches quickly, safely and autonomously and determine if the system is reliable.

“Heterogenous data from multiple Internet of Things (IoT) devices and from multiple sources risk extra data being inserted. It can be noisy and it can even be a security threat,” said Gokaraju.

“We are applying adversarial machine learning by taking data-applying statistical learning models to predict malicious attacks for future events, or a new sample of data,” said Roy.

The researchers are taking AI to the next level with explainable AI which allows the work to prepare for future events by adding noise to the models. This practice builds resilience into the systems. The grant will support one student who will take a headshot photo of a person and create 1,000 fake versions of the image to train the AI models to recognize imposter data and decipher which photo is authentic. In some deep fake videos, the team processes an astounding 200 terabytes of data to prove or disprove authenticity.

A&T is designated by NSA as a National Center of Academic Excellence in Cybersecurity for research and education – two out of three of the agency’s possible designations, with cyber options being third. A&T is the first historically Black college or university to hold two designations.

“One of the strengths of the computer science department and the COE is cybersecurity. We excel in evolving cybersecurity from an AI perspective,” said Roy.

Media Contact Information: uncomm@ncat.edu

Headlines, College News, All News