Nathan Kahl

George �鶹��Ƶ and King’s College London win Google Award to fight AI-driven romance scams targeting older adults

Nathan Kahl — Sun, 26 Oct 2025 23:23:41 +0000

George �鶹��Ƶ and King’s College London win Google Award to fight AI-driven romance scams targeting older adults Nathan Kahl Sun, 10/26/2025 - 19:23

The term “catfishing” entered mainstream use with the release of the eponymous 2010 documentary investigating the practice of creating fake online personas, frequently to scam someone. Today, AI is pushing this practice to a frightening new level, with the prospect of older adults falling for what are called “romance scams.”

Sanchari Das, an assistant professor in �鶹��Ƶ’s Department of Information Sciences and Technology, and Ruba Abu-Salma, an associate professor in the department of computer science at King’s College London, received a $100,000 Google Academic Research Award to help keep older adults safe online. Their project, Scammed by Love: Toward Mitigating AI-Facilitated Romance Scams Against Older Adults Through Empirical and Cross-National Studies, will investigate the prevalence of such scams targeting older adults and will create educational tools to help them safely navigate dating apps.

The researchers will conduct a multi-national study spanning 13 countries, including the United States and the United Kingdom. Das, who has published research on older adults and their online dating habits, said, “We think about older adults in terms of health care needs, but we don't think about dating and romance. I learned that a majority of older-adult scams come from scam dating, and the victims cannot even share that it happened with their support networks, like their children, because it would be such a taboo and embarrassing thing to share.” Das added that an estimated $3 billion is lost each year in the United States to romance scams across all age groups, based on reported figures. “And that’s only what’s being reported.” Das is concerned that while romance scams are among the most prevalent forms of fraud affecting older adults, this population remains understudied in this context.

AI is making romance scams even harder to detect. iStock photo.

With roughly 12 percent of Americans over 70 having used a dating app or site, there is significant exposure for financial harm. Given the rapid advances in AI-generated images, video, audio, and simulated emotional expression, these attacks are increasing. One news story (of many) tells of a UK resident who scammed at least 10 female victims over 11 years out of 100,000 British pounds.

Abu-Salma, who has also published research on older adults and other at-risk groups, noted that according to a recent 2025 study by Lloyds Bank, “Individuals aged 55 to 64 are the most targeted age group, accounting for 20 percent of romance scam victims, followed by those aged 65 to 74, who represent18 percent of cases.”

In response to these risks, Das and Abu-Salma’s project will proceed in three phases. First, they will analyze data from focus groups and surveys of older adults in countries with high dating-app adoption. Second, they will develop educational safety materials. Third, they will hold in-person community workshops in the United States and the UK, partnering with older adult communities to refine and disseminate these materials. The team will work with organizations worldwide to share their findings online.

Das said she is excited about this collaboration and grateful to receive such highly competitive Google Academic Research Award in the Trust, Safety, Security, and Privacy research domain, noting, “We all know about deepfakes, but the bigger issue is that scammers are no longer just swapping faces. They are creating entirely new personas that feel real, so you may not even question whether the person you are talking to exists.” She added, “Our goal is to equip older adults with simple tools and support so they can connect online with confidence.”

In This Story

Sanchari Das

Topics

Cyber Security

Information Sciences and Technology Departmenat

Older Adults

Research

Google

Artificial Intelligence

Anonymity’s ARX nemesis

Nathan Kahl — Mon, 20 Oct 2025 15:12:08 +0000

Anonymity’s ARX nemesis Nathan Kahl Mon, 10/20/2025 - 11:12

A team of faculty and students from �鶹��Ƶ recently discovered a vulnerability in a widely used anonymization tool. They presented their findings last week in Taiwan at the Association for Computing Machinery Conference on Computer and Communications Security (ACM CCS), one of the world’s most prestigious computer security conferences, with a very low paper acceptance rate.

The problem they discovered was with ARX, an open-source data anonymization tool, which provides what is referred to as k-anonymity. It is a commonly used tool in clinical settings to maintain data privacy. The discovery is particularly important at a time when more and more private data is being used in a variety settings and systems.

Evgenios Kornaropoulos, an assistant professor in the Department of Computer Science, said, “Sometimes this type of microdata is very valuable. Tools like ARX anonymize the data so it’s compliant with HIPAA, and then that data can be shared with policymakers, engineers, scientists, and others who will make policy decisions and scientific studies based on this information.”

While many people receiving medical care are comfortable with their information being included in larger data sets for purposes of medical discovery, they likely do not want personal, identifiable information released.

“For patients, and especially those from more vulnerable or marginalized communities, data privacy is not an abstract concern, it’s deeply tied to trust in the health care system," said paper co-author Rebecca Sutter, a professor in the School of Nursing, and director of the MAP Clinics. "The integrity of tools like ARX matters because they underpin how we protect patient information while advancing public health research. When anonymization fails, it’s not only a technical vulnerability, it’s a human one."

Computer science PhD student Somiya Chhillar, lead author on the paper, said that the ubiquity of ARX was a reason they chose to study it. “We wanted to see how ARX really operates and how the algorithm works. While we were doing that, we realized that a few of the steps that it takes are very opportunistic, and because of that, it leaks information that we shouldn't be finding out by just looking at the anonymized data.”

Chhillar said that ARX follows a “greedy strategy,” explaining that there is a tension between privacy and utility. When data is shared, it should be useful without leaking information. The best metric for deciding the data usefulness is measured by information loss while the data is being made private. “The algorithm tries to minimize information loss while anonymizing the data, and ideally, we want the least possible information loss. And that's the greedy aspect of it; it's not always going for the most private, it’s just going for the most utility.”

This is what backfires, said Kornaropoulos, because experts (or attackers) can reverse engineer the anonymization steps the algorithm took throughout its execution to maximize utility and figure out properties of the data that were there before and after this anonymization step.

Kornaropoulos praised his student’s dedication to the discovery, noting that Chhillar “worked on this project for a while, and this code base of ARX is an elaborate tool, with thousands of lines of code. She had to go through it to truly internalize understand everything that's going on there,” he said.

The project was supported by a Commonwealth Cyber Initiative (CCI) grant from the program, “Securing Interactions between Humans and Machines,” and as a requirement of the grant, the project crossed different parts of the university. The College of Engineering and Computing collaborated with �鶹��Ƶ and Partners (MAP) Clinics, which provided the data.

In This Story

Evgenios Kornaropoulos

Rebecca Sutter, DNP, APRN, BC-FNP

Learn more about Cybersecurity at George �鶹��Ƶ

Topics

computer science

Cyber Security

Data Mining in Health Care

Commonwealth Cyber Initiative (CCI)

Research

Resilient robots in rough terrain

Nathan Kahl — Tue, 14 Oct 2025 01:01:54 +0000

Resilient robots in rough terrain Nathan Kahl Mon, 10/13/2025 - 21:01

Xuesu Xiao oozes enthusiasm when he talks about his off-road robotics testbed in �鶹��Ƶ’s new Fuse building at �鶹��Ƶ Square.

The Robotixx team provided demonstrations to visitors during the Fuse grand celebration in September. Photo by Ron Aira/Office of University Branding

“I'm extremely excited about this because it’s an only-of-its-kind thing on the planet. Nowhere else has such abilities. Off-road mobility and autonomy is a very difficult area to study and investigate because you need to have supporting facilities, you need to have off-road terrain, and you need to have the vehicles. We have all of that,” said the computer science assistant professor.

Xiao is a roboticist who develops algorithms to make robots not only capable, but also intelligent. His goal is that robots will someday work in the real world; humans will stand aside while robots go places that humans don't want to—or simply cannot—go.

“We want them to go to a forest to search for a victim who is lost in the woods,” he said. “We have a proposal in review right now to build robot firefighters. We want them in the field, doing things like scientific exploration in the South Pole.”

Presently, autonomous driving generally involves road networks with traffic lights and lane markings. Xiao’s research focuses on motion planning: using machine learning to determine a series of movements to get a robot from a starting point to a goal while avoiding obstacles.

“That is where AI comes into place,” said Xiao. “We want robots not only to be programmed to do things, but also to learn from themselves, from the humans around them, and from each other.”

Xiao hosted visitors from the RoSys Group of the University of Catania, Italy. Pictured: assistant professor Giuseppe Sutera, Xiao, PhD student Francesco Cancelliere, and assistant professor Dario Calogero Guastella in the test bed. Photo by Ron Aira/Office of University Branding

The testbed where Xiao and students put their robots through their paces is about 700 square feet of rocks and faux vegetation. Xiao is intimately familiar with the space, having placed most of the rocks himself. “I'm very, very proud of this testbed, which has more than seven tons of rocks. I took five of our students of seven or eight trips to the landscaping store. It was an exhausting week!”

The terrain is rugged, designed to represent a real-life situation. “Off-road environments may get the robot stuck because the terrain may be deformable—changing shape under weight,” Xiao said. “Off-road terrain may make the robot roll over because, if there are a lot of rocks, they may need to crawl over them. You don't want the vehicle to roll over on top of those things. Successfully maneuvering this terrain is a very, very big thrust of our lab.”

The testbed is built to a one-tenth scale. A motion-capture system of eight cameras (soon to be upgraded to 16) surveils the entire terrain. Xiao says that his team knows exactly how the vehicles are moving with six-degrees-of-freedom tracking.

The year has been busy for Xiao, who in 2025 won the university-wide Presidential Award for Faculty Excellence and the College of Engineering and Computing’s Excellence in Research Award. He is also George �鶹��Ƶ’s sole nominee for the Rising Star Outstanding Faculty Award by the State Council of Higher Education for Virginia.

While he appreciates that recognition, Xiao isn’t one to rest on his laurels. He is particularly excited about what’s next in the lab: He just ordered an autonomous Polaris “side-by-side”—a $300,000 autonomous off-road vehicle that can hold two people. He also talked about a capability that allows the testbed to be used far from George �鶹��Ƶ.

“I was at a conference in Korea, and we opened our testbed to the entire world by using a web interface. So, the attendees of the Conference on Robot Learning in Korea could drive our robots in Fuse in Arlington, Virginia.”

With robots at George �鶹��Ƶ taking commands from the other side of the globe, Xiao is showing that in robotics, there are no borders—only new frontiers to explore.

Explore AI at George �鶹��Ƶ

In This Story

Xuesu Xiao

Topics

Autonomous Systems

Fuse at �鶹��Ƶ Square

robotics

Artificial Intelligence

Research

�鶹��Ƶ Autonomy and Robotics Center

A one-woman team puts many eyes in the skies

Nathan Kahl — Wed, 01 Oct 2025 19:37:27 +0000

A one-woman team puts many eyes in the skies Nathan Kahl Wed, 10/01/2025 - 15:37

Fatima Majid, a �鶹��Ƶ senior majoring in cyber security engineering, was not just the only one-person team in the top 10 award winners at a recent National Defense Industrial Association (NDIA) cyber competition, she was the only student team. Majid placed ninth out of 51 teams, most of them comprising industry experts.

“I went in, and they were all professional teams, like from Lockheed Martin. I thought, ‘I want to go home,’” said Majid with a laugh, describing her initial cold feet. “But I told myself I could do it. It helped that it was hosted at George �鶹��Ƶ, and I had professors there giving me support.”

Her project focused on how the Department of Defense (DoD) can protect critical U.S. infrastructure against low-cost drone attacks at scale, informed by Ukraine’s “Operation Spiderweb,” which used 117 drones to attack Russian air bases in June 2025.

Majid spent the summer in Richmond during a public policy internship. Photo provided

Majid’s lightbulb moment came with a flash as bright as a Virginia speed camera catching a lead-footed driver. Considering the significant network of traffic cameras in the commonwealth, she conceived SkyEyes. This applies an artificial intelligence (AI) model to the live feed of the Virginia 511 camera network, which provides real-time traffic information to citizens and transportation officials. SkyEyes demonstrated how a low-cost, AI-enabled surveillance layer could differentiate threats from non-threats, employing geofencing logic to define safe versus threat zones around sensitive sites.

“I understand how drones work because of what I’ve done at George �鶹��Ƶ’s MIX lab—and since I know how to build it, I also know how to jam it. I trained the AI model on a data set provided from a contest sponsor, and that data set had drone imaging and drone prototyping,” she said. “The camera feeds can find objects flying, but what if it’s an Amazon drone, for example? Then I added geofencing and threat analysis to observe the behavior of the drone—if it's a drone at 2 a.m., for example, maybe that's sketchy. So, the model gets smarter.”

Majid said that to access the cameras, all she had to do was make a phone call to the right person and explain her project. She cited time spent this summer at the Virginia Academy of Science, Engineering, and Medicine in the Undergraduate Policy Program (VASEM UPP) in Richmond as giving her confidence and exposure to how government works.

K. L. Akerlof, an associate professor in the Department of Environmental Science and Policy at George �鶹��Ƶ, said, “The VASEM UPP is a unique opportunity for undergraduates to learn about opportunities in science policy at the state level. The immersive experience of spending a week in Richmond visiting the General Assembly and state agencies, while getting a crash course in how research evidence relates to public policy, can open new doors and career pathways.”

Majid said the strong showing gave her tremendous exposure to influential professional contacts. She fielded several questions about her simulation and future professional plans from a man she only later realized was Retired Brigadier General and NDIA Executive Vice President Guy Walsh.

“Because he showed interest, after he walked away, a crowd of people gathered around to ask me questions. It was very validating.”

She also had a long conversation about her project with Harley Stout, acting chief digital and AI officer at the Joint Chiefs of Staff.

Majid is still pleasantly shocked by her top-10 finish. She is working with Mohamed Gebril, an associate professor in Cyber Security Engineering, on expanding the research. She is confident that such a cost-efficient solution for critical infrastructure protection against drone attacks will attract more funding opportunities.

She credited the supportive culture at �鶹��Ƶ—and her family—for keeping her grounded and encouraging her throughout, saying their support made the accomplishment even more meaningful.

In This Story

Mohamed Gebril

Topics

Cyber Security Research

Department of Cyber Security Engineering

competitions

Drones

Students

Artificial Intelligence

The MIX

Defense and Security

CEC students

NSF grant supports nanofabrication workforce training

Nathan Kahl — Sat, 13 Sep 2025 22:43:45 +0000

NSF grant supports nanofabrication workforce training Nathan Kahl Sat, 09/13/2025 - 18:43

�鶹��Ƶ faculty members received nearly $1 million in a grant from the National Science Foundation (NSF) to establish a program providing hands-on nanofabrication training and education to people seeking a career change.

The funding comes under NSF’s ExLENT Pivots track and will serve the national interest by expanding experiential learning pathways in emerging hardware technologies driven by nanofabrication. It focuses on increasing workforce opportunities in Northern Virginia, a region positioned for growth in emerging technology sectors, such as artificial intelligence hardware, quantum information science, data center engineering, and sustainable manufacturing.

Ahn's research focus is nanofabrication. Photo by Office of University Branding

“The most important term is ‘experiential,' and it’s rare for NSF to provide this kind of funding," said Ethan Ahn, an associate professor in the Department of Electrical and Computer Engineering and a founding director of �鶹��Ƶ Nanoelectronics Laboratory, who is the lead principal investigator (PI) on the project. "The idea here is to produce this workforce for very high tech-focused areas by applying the learning-by-doing model.”

George �鶹��Ƶ is already known as one of the commonwealth's largest producers of tech talent.

The program provides an opportunity for working professionals who are looking to change careers; even those with no STEM background are encouraged to apply, as suggested by the “pivot” track from NSF.

“We’ll take people regardless of their background, education, or readiness," said Ahn. "We’ll provide everything they need to be job-ready by training participants with both lab and soft skills, providing connections with industry, and, if hired, paying them during their summer employment."

Co-PI Jessica Rosenberg, professor of physics and director of education at George �鶹��Ƶ’s Quantum Science and Engineering Center, said, “Nanofabrication is important for a range of high-tech industries, including semiconductors, biomedical, and energy. We’re creating a program to provide the support, education, and training needed for individuals to get a job in this growing sector. It will help them meet their goals while also supporting a regional talent pipeline.”

The program will run for three years, with approximately 20 participants every year. A key component of the initial phase of the program is access to a mobile app used for virtual training. The team is developing the app because they recognize that participants may have existing responsibilities that make coming to campus a challenge. After participants complete this first phase on their own schedule, they will then go to campus to participate in a research lab for the hands-on portions of the program.

Jessica Rosenberg is director of education at George �鶹��Ƶ's Quantum Science and Engineering Center. Photo by Office of University Branding

Ahn said a particularly interesting element of the program is a guaranteed interview in the industry. They are already working with BAE Systems, General Dynamics, and Northrop Grumman and expect this number of companies participating to grow through a collaboration with the Virginia Economic Development Partnership. "The hope is that the participants will be able to turn the opportunity into a full-time position,” he said.

George �鶹��Ƶ will give program participants microcredentials, competency-based qualifications indicating a mastery of a subject, as they complete each phase of the program. This opportunity’s timing is ideal for Virginians looking to make a career change and supports the university’s efforts in building Virginia’s high-tech workforce.

Ahn was inspired to apply for the funding in part by the fact that the new Innovation District, led jointly by �鶹��Ƶ, Prince William County, and the City of Manassas, and anchored by George �鶹��Ƶ’s Science and Technology Campus, co-locates the university with industry and government to create a dynamic collaboration model.

“Relationships between university and industry are common, but this also includes government,” said Ahn. “We already have this established network via the Innovation District and in my proposal I said we are going to focus on nanofabrication, which is my area of expertise. I think those things made it appealing to NSF.”

Jill Nelson, professor and associate dean for undergraduate programs, and Nathalia Peixoto, associate professor, both in the department of Electrical and Computer Engineering, are also PI's on the grant.

The program will welcome its first cohort in spring 2026.

In This Story

Ethan Ahn

Learn more about Research at George �鶹��Ƶ

Topics

Nanofabrication Facility

Innovation District

National Science Foundation

Micro-credentialing

Quantum Science and Engineering Center

Research

College of Science

Science and Technology Campus

workforce

College of Engineering and Computing

Dialing up new 9-1-1 training with AI

Nathan Kahl — Tue, 09 Sep 2025 18:34:13 +0000

Dialing up new 9-1-1 training with AI Nathan Kahl Tue, 09/09/2025 - 14:34

�鶹��Ƶ’s Center for Resilient and Sustainable Communities (C-RASC) is working with the Fairfax County Department of Public Safety Communications (DPSC) to implement artificial intelligence (AI) and possibly forever change the way 9-1-1 operators are trained.

Photo provided

Funded by a grant from the National Institute of Standards and Technology (NIST) through a congressional earmark led by the late congressman Gerry Connolly, C-RASC started working with Fairfax’s DPSC in 2024. The goal was to implement AI to improve the resiliency of the county’s emergency response system, using AI to assist 9-1-1 operators by answering calls.

But Kathy Laskey, professor emerita of Systems Engineering and Operations Research, said, “There are all kinds of legal and other sticky issues with putting AI on the phone with people calling 9-1-1.” The researchers pivoted quickly. “The DPSC told us that where they could use help is training, because operators need a lot of practice in taking calls before they actually respond to real callers.”

Training is very labor-intensive in a 9-1-1 call center, and typically an instructor will play the role of a 9-1-1 caller and improvise a call scenario from a large scenario booklet, with the trainee responding and getting feedback on the responses they give.

The Conversational AI-assisted Scenario Training System developed by the George �鶹��Ƶ faculty together with industry partners and guided by DPSC is very sophisticated in generating realistic scenario calls and is customizable.

“Trained in part on the scenario booklet being fed into this AI system, the scenario narration by the voicebot is done in such a way that it mimics a wide range of typical caller behavior,” said Hemant Purohit, associate professor in the Department of Information Sciences and Technology and director of the Humanitarian Informatics Lab.

From left, Hemant Purohit, Monika Bochert of NIST, and Lin Wells impressed the Public Safety Innovation Summit Conference attendees with the new technology. Photo provided

Lin Wells, executive advisor to C-RASC, added, “Each caller may have a different situation, and the training staff helped us identify those variables in terms of the components they want trainees to practice.”

The project team is enhancing the system in several ways. One is by incorporating various difficulty levels so the system can be used both for new employees and as a refresher for existing staff. Another is to add realism, such as the sounds of passing cars in a scenario involving a traffic accident. Another enhancement is to include diverse speech patterns, such as accents, different tones of voice, language, and other real-world challenges the operators face.

Purohit said the team recently attended the NIST Public Safety Innovation Summit Conference (5 x 5) in Seattle and got a very positive response to the technology. A 9-1-1 communications and training coordinator from another state, who had 27 years of work in public safety and received this year’s award at the conference, said the technology was the best she had seen.

Laskey added that Fairfax County also has been very supportive of the work. “I taught at George �鶹��Ƶ for 32 years, and I've worked on a lot of research projects. I've never before in my life demonstrated a software project to a client where the recipients applauded when we were done.”

Perhaps the most important endorsement of all came from the 9-1-1 training instructors, who were impressed with how well the AI callers mimicked real-life situations and how this could save them hours of training.

The other project team members include Paul Houser, professor and executive director of C-RASC, and Karina Korostelina, professor and director of the Sustainable Peace Lab at the Carter School for Peace and Conflict Resolution. Korostelina is leading the project’s community engagement study, which has involved interviews with Fairfax County citizens. Her findings showed the barriers to effective 9-1-1 services include language, delays with interpreter access, limited public understanding of emergency-vs.-non-emergency use, rigid questioning protocols, human error, and the need for more experience in handling mental health crises.

“The increase in adaptive capacity of new technologies requires deep considerations for micro-dynamics of identity and power in communities,” said Korostelina. “To build this capacity as a part of community resilience, the project will codesign the systemic approach for community engagement, including public education campaigns using schools, libraries, media, and community workshops to teach the difference between emergency and non-emergency calls and how to conduct emergency calls more efficiently.”

The project team is supported by doctoral students Ramya S. Nayaka and Hamad Edhah Yaslam Aldhlea from the College of Engineering and Computing and the Carter School for Peace and Conflict Resolution respectively, as well as industry partners OmniBot Advisors and BlueHorse Digital.

In addition to working with Fairfax County on a possible continuation of the project the group is exploring academic-industry partnerships and preparing to discuss possibilities for commercialization with the university’s Office of Technology Transfer.

Explore AI Research at George �鶹��Ƶ

In This Story

Hemant Purohit

Karina Korostelina

Topics

AI and human factors

Emergency Management

Carter School for Peace and Conflict Resolution

Center for Resilient and Sustainable Communities

National Institute of Standards and Technology NIST

Research

Artificial Intelligence

Bluetooth blues? George �鶹��Ƶ turns a bug into a big win

Nathan Kahl — Thu, 21 Aug 2025 01:28:32 +0000

Bluetooth blues? George �鶹��Ƶ turns a bug into a big win Nathan Kahl Wed, 08/20/2025 - 21:28

�鶹��Ƶ College of Engineering and Computing faculty members made a significant impact at the 2025 USENIX Security Conference, in Seattle, including winning the Distinguished Artifact Award.

Photo provided

“Tracking You from a Thousand Miles Away! Turning a Bluetooth Device into an Apple AirTag Without Root Privileges,” a paper by the Department of Computer Science’s PhD student Junming Chen, PhD candidate Xiaoyue Ma, and Lannan Luo and Qiang Zeng, both associate professors, won the prestigious Distinguished Artifact Award, which was granted to only six of the 2,385 submitted papers.

Their research demonstrated how to turn Bluetooth devices into a trackable "AirTags" without requiring root privileges. The research revealed a vulnerability in Apple's Find My Network that can leverage more than a billion active Apple devices as unwitting finders. The high-quality artifact with its complete documentation and practical implementation provides researchers with an invaluable toolkit for evaluating tracking vulnerabilities in crowd-sourced location networks. Their work was previously reported by the university and received media attention in various outlets.

A crowd gathers at the team's poster session. Photo provided.

USENIX Security is one of the country’s premier cybersecurity conferences, and this year received 2,385 paper submissions with a 17% acceptance rate. The following George �鶹��Ƶ research papers were also accepted at USENIX this year. 

Rowhammer-Based Trojan Injection: One Bit Flip Is Sufficient for Backdooring DNNs
Xiang Li, Ying Meng, Junming Chen, Lannan Luo, and Qiang Zeng, �鶹��Ƶ

DISPATCH: Unraveling Security Patches from Entangled Code Changes
Shiyu Sun and Yunlong Xing, �鶹��Ƶ; Xinda Wang, University of Texas at Dallas; Shu Wang, Palo Alto Networks, Inc.; Qi Li, Tsinghua University; Kun Sun, �鶹��Ƶ

TYPEPULSE: Detecting Type Confusion Bugs in Rust Programs
Hung-Mao Chen and Xu He, �鶹��Ƶ; Shu Wang, �鶹��Ƶ and Palo Alto Networks Inc.; Xiaokuan Zhang and Kun Sun, �鶹��Ƶ

LLMmap: Fingerprinting for Large Language Models
Dario Pasquini, RSAC Labs; Evgenios M. Kornaropoulos and Giuseppe Ateniese, �鶹��Ƶ

Explore the College of Engineering and Computing

In This Story

Qiang Zeng

Topics

A calibration from campus to the cosmos

Nathan Kahl — Fri, 15 Aug 2025 14:48:15 +0000

A calibration from campus to the cosmos  Nathan Kahl Fri, 08/15/2025 - 10:48

In This Story

Peter Pachowicz

�鶹��Ƶ is the home of the $20 million Landolt NASA Space Mission, placing a satellite that will serve as an artificial “star” in orbit around the Earth…and College of Engineering and Computing students are instrumental in the payload’s construction. It’s experience that not only significantly bolsters a resume, but also allows students to work on a project of their dreams.  

Fatima Bahzad, a computer science graduate student said, “It’s the most fun I’ve ever had in school.”  

Bahzad said that working on the mission is the most fun she's had at George �鶹��Ƶ. Photo provided.

Bahzad explained her role: “The satellite, which is about the size of a bread box, is comprised of layers and each layer has a chip and the chips need to talk to each other. There’s also a primary chip that sends commands to all of them. I’m working on the communication between the chips.”

Getting an opportunity to turn theory into practice is incredibly gratifying, she said. “Learning theoretical stuff is one thing, but applying it is another. This is one of the first opportunities I’ve had to work on something ‘real.’” 

The payload is being built in partnership with the National Institute of Standards and Technology (NIST), with whom the teams checks in frequently. The artificial star will allow scientists to calibrate telescopes and more accurately measure the brightness of stars ranging from those nearby to the distant explosions of supernova in far-off galaxies. By establishing absolute flux calibration, the mission will address several open challenges in astrophysics, including the speed and acceleration of the universe expansion. 

On the engineering side, the project is led by Piotr Pachowicz, an associate professor of electrical and computer engineering. Pachowicz brought PhD student Jay Deorukhkar, one of his advisees, onto the project.  

Deorukhkar said he’s been working on testing the individual payload modules. These will later integrate with the satellite bus, which is the core structure and support system that is separate from the specific payload. Reliability of small satellite buses against radiation effects is also the subject of his dissertation, making this work particularly relevant.  

Deorukhkar is working on his second project that will send an object into space. Photo provided.

“My role as an engineer is to help design, prototype, and test specific payload components. We’re early in the design stage, working towards the preliminary design review of our payload.” He added, “The great thing about this project is really the end-to-end experience, from the inception of the mission  to the actual launch and operation of the science payload.”  

Deorukhkar has some experience with that, having worked on the ThinSat project in 2021, which sent a small satellite into low earth orbit. He knows about the anxiety as the launch date nears. “It's a bone-chilling moment when the rocket starts and you know your device is on there and it’s out of your hands at that point. You can only pray and trust that everything you've learned and you've done just works.”  

Bahzad is still trying to keep her feet planted on the ground, so to speak, at times overwhelmed to be part of something so significant. “We met recently with some people from NASA and I was just sitting in the corner and couldn’t believe it. I was just thinking, ‘This is so cool!’”   

Housed in the George �鶹��Ƶ College of Science, the mission launches in 2029, when the artificial star will orbit earth 22,236 miles up, far enough away to look like a star to telescopes back on Earth. This orbit also allows it to move at the same speed of the Earth’s rotation, keeping it in place over the United States during its first  year in space. 

�鶹��Ƶ faculty and students will work together with nine other organizations beyond NASA and NIST on this first-of-its-kind project for a university in the Washington, D.C., area. 

Topics

Space

Electrical and Computer Engineering

satellite

NASA

National Institute of Standards and Technology NIST

Students

One tiny flip that opens a backdoor in AI

Nathan Kahl — Mon, 11 Aug 2025 13:13:32 +0000

One tiny flip that opens a backdoor in AI Nathan Kahl Mon, 08/11/2025 - 09:13

A self-driving motor vehicle is cruising along, its numerous sensors and cameras telling it when to brake, change lanes, and make turns. The vehicle approaches a stop sign at a high rate of speed, but instead of stopping, it barrels through, causing an accident. The problem will probably never be found by investigators, but instead of reading the stop sign as a stop sign, the car had been hacked to see it instead as a speed limit sign.

According to research by �鶹��Ƶ’s Qiang Zeng, an associate professor in the Department of Computer Science, and PhD student Xiang Li and colleagues, it is remarkably simple for a would-be hacker to pull off such a feat.

“An attacker can selectively flip only one bit, and this changing of the bit from 0 to 1 allows an attacker to attach a patch onto any image and fool the AI system. Regardless of the original image input, that patched image will be interpreted as the attacker’s desired result,” said Zeng.

So, if the hacker wants an artificial intelligence (AI) system to see a stop sign as something else, or a cat as a dog, the effort is minimal. Consider a scene potentially pulled from a Mission: Impossible movie, where a corporate spy can pass himself off as a CEO, gaining access to sensitive information.

Zeng and colleagues presented a paper with the findings at USENIX Security 2025, one of the nation’s premier cybersecurity conferences.

AI systems have what’s called a deep neural network (DNN) as a key component. DNNs let AI handle complex data and perform many different tasks. They work by using numerical values, called weights, each typically stored in 32 bits. There are hundreds of billions of bits in a DNN, so changing only one is particularly stealthy, according to Zeng.

“Once the attacker knows the algorithm, then it can take literally only a couple minutes to make the change. And you won’t realize you’ve been attacked because the AI system will work as usual. Flipping one bit effectively sneaks a backdoor into AI, exploitable only by those who know the patch,” he said.

Prior work in this area typically added a patch tailored to the original image—for example, modifying a stop sign specifically so that it is misclassified as a 65 mph speed limit sign. This new research uses what’s called a uniform patch that works regardless of the original input; the hacker could cause the system to interpret various signs as a speed limit sign. This input-agnostic attack represents a newer and more dangerous threat.

When they began the project the researchers wanted to learn the minimum level of effort needed to launch such an attack, recognizing that flipping hundreds of bits is impractical and gets exponentially difficult. “It turned out, we only needed to flip one,” Zeng said with a laugh. The team named their attacking system, appropriately, OneFlip.

The researchers are now only looking at the implication for images, as image classifiers are among the most popular AI systems, though they suspect this hacking technique could also work for things like speech recognition. Zeng said that their success rate during testing was near 100% and stressed that all DNN systems are likely to be subject to such hacking.

This does not necessarily mean such hacking will run rampant. To launch the attack, Zeng said, there are two requirements: access to the exact weights (numerical values that the model learns during training of the AI system) and the ability to execute code on the machine hosting the model. For example, in cloud environments, attackers might exploit shared infrastructure where multiple tenants’ programs run on the same physical hardware.

Learn about AI at George �鶹��Ƶ

Topics

Cyber Security

Cyber Security Research

computer science

Autonomous Cars

Artificial Intelligence

Research

I-Corps in 3D: Streaming toward better telehealth

Nathan Kahl — Thu, 17 Jul 2025 14:19:57 +0000

I-Corps in 3D: Streaming toward better telehealth Nathan Kahl Thu, 07/17/2025 - 10:19

In This Story

Bo Han

A team in �鶹��Ƶ’s Computer Science Department is taking part in the National Science Foundation’s (NSF) I-Corps program to study the effectiveness of 3D streaming in telehealth environments.

Bo Han, an associate professor of computer science and expert in computer networks and security and privacy, received $50,000 for his project, Translation Potential of Next Generation Telepresence Enriched by Immersive Technologies. Nan Wu, the entrepreneurial lead on the team, said, “The idea is to build a system for immersive telepresence over the internet. It’s essentially a 3D video stream but the technical improvement is how it saves bandwidth by using user-motion prediction and content sampling, improving the quality of the experience.”

3D telehealth services may help remote patients in urgent need of care. iStock photo.

Wu explained that the technical adjustment is made based on the receiver’s movements. “Since I’m looking at the front side of you, the streaming only needs to send me the front side, not the back. I don’t need to see all around you, even though it’s 3D. Moreover, when you’re farther away, a lower-resolution stream is enough since I won’t notice the difference. And if you’re hidden behind other content, there’s no need in sending what I can’t see.”

I-Corps is, according to the NSF website, “an immersive, entrepreneurial training program that facilitates the transformation of invention to impact. This seven-week experiential training program prepares scientists and engineers to extend their focus beyond the university laboratory—accelerating the economic and societal benefits of NSF-funded and other basic research projects that are ready to move toward commercialization.”

Since the program began in 2012, I-Corps has ushered 2,500 teams, and nearly 1,400 of them launched startups that cumulatively raised $3.16 billion in subsequent funding.

A portion of the I-Corps mentoring is devoted to customer discovery. Teams must interview at least 100 potential customers to determine market interest. According to Wu, the team didn’t find a robust market for using the technology the way they initially envisioned because staff shortages is the biggest pain point in the field, not technical shortcomings. “We thought nurses could monitor patients in 3D and save time, but there are many AI tools already doing that, monitoring 40 patients at once,” he said.

Wu said they’ve shifted to using the technology emergency situations, believing the tools will be useful in high-pressure trauma care or emergency cases. “We are thinking of a situation where there is a remote expert connected to a patient in a rural area, for example, where there is no expert on site.” Wu said that more interviews are needed to determine the market demand for such use.

Topics

Telehealth

I-Corps

National Science Foundation

immersive technologies

Research

Nathan Kahl

George �鶹��Ƶ and King’s College London win Google Award to fight AI-driven romance scams targeting older adults

In This Story

Related Stories

Topics

Anonymity’s ARX nemesis

In This Story

Related Stories

Topics

Resilient robots in rough terrain

Related Stories

In This Story

Topics

A one-woman team puts many eyes in the skies

In This Story

Related News

Topics

NSF grant supports nanofabrication workforce training

In This Story

Related Stories

Topics

Dialing up new 9-1-1 training with AI

In This Story

Related News

Topics

Bluetooth blues? George �鶹��Ƶ turns a bug into a big win

In This Story

Related News

Topics

A calibration from campus to the cosmos

In This Story

Topics

One tiny flip that opens a backdoor in AI

Related Stories

Topics

I-Corps in 3D: Streaming toward better telehealth

In This Story

Topics