UMD CS News

When AI Goes on Defense

Samuel Malede Zewdu — Thu, 10 Sep 2026 15:21:42 +0000

CS Ph.D. student Manan Suri and Professor Dinesh Manocha are part of a research team developing CyberForge, a framework that generates realistic vulnerabilities to train AI agents for software defense.

September 10, 2026

For a cyberattacker, one overlooked weakness can be enough. For a defender, the challenge is broader: find weaknesses and fix them before attackers exploit them.

That imbalance is becoming more consequential as artificial intelligence agents improve at writing and analyzing code. , Manan’s recent work on (NIST),

“AI agents are getting stronger at writing code, and as expected, I think that creates a lot of challenges for cybersecurity,” Suri said.

Uneven playing field

AI-assisted coding has increased the speed and scale at which software can be analyzed, raising the stakes for security teams trying to spot weaknesses before attackers exploit them.

“An attacker in a cybersecurity scenario only needs to find one vulnerability to get into your system, whereas a defender ideally needs to find all of them and fix all of them,” Suri said.

AI agents add another dimension to that imbalance. Unlike human operators, they can run continuously, handle multiple tasks at once and scale across large amounts of code. For Suri and his collaborators, that raised a practical research question: whether similar capabilities could be directed toward identifying and repairing weaknesses before they are exploited.

“It’s this asymmetry which creates this race, a race where we must empower defenders as much as we can,” he said.

Data ceiling

Building capable defensive systems depends in part on what they can learn from. In cybersecurity, much of that material comes from records of vulnerabilities that researchers, companies or developers have already discovered and disclosed.

Those examples are useful because they come from actual software, but they represent a finite historical record.

“At one point, you’re going to run out of the historically available data,” Suri said.

The problem becomes more pronounced when a company wants to train a model on its own software. A project may have relatively few documented vulnerabilities, leaving too little material to train an AI system around the specific structure and behavior of that codebase.

“Very often, I would imagine a scenario where an enterprise wants to train a cybersecurity agent that’s specific to their own project,” Suri said. “Individual projects don’t have enough exposed vulnerabilities.”

Synthetic data offers one way to expand that pool, but artificially created weaknesses don't necessarily reflect how vulnerabilities emerge in working software. CyberForge was developed to generate examples using information already contained within a real project.

Building CyberForge

This approach works with open-source C and C++ repositories and looks for places where code can be altered to introduce a security weakness without breaking the software’s normal behavior.

To create a useful training example, CyberForge must introduce a vulnerability without stopping the software from working normally. The researchers can then use a specific input to trigger the weakness and show the model the difference between secure and vulnerable code.

“Given limited historic data, you use the signals from the project to create realistic synthetic vulnerabilities,” Suri said. “What this allows the agent to learn is how to defend itself in case of a real vulnerability.”

The project emerged during Suri’s spring semester as a Professional Research Experience Program fellow at NIST. His background focused on AI agents, while team members brought experience in cybersecurity datasets, vulnerability analysis and standards.

The research brought together researchers from NIST, UMD,

“I got to work with people who came from a background of cybersecurity,” Suri said. “This is a group of people who have released major datasets and reports and standards for vulnerabilities.”

That combination of expertise also affected how the team evaluated possible research directions. Suri said subject-matter expertise helped distinguish technically plausible ideas from approaches that would be less useful in practice.

“When you’re solving a domain-specific problem, the value that experts bring to that problem is immense,” he said.

Measuring defense

The researchers then tested whether training on CyberForge-generated vulnerabilities made the models better at protecting software. Across several experiments, performance increased by about 3.3 to 15 percentage points, Suri said.

“We wanted to see where open-source models stood on cybersecurity tasks and whether training them on this new data could improve their performance,” he said.

On one benchmark, a model successfully defended against vulnerabilities about 58% of the time before CyberForge training. After training, that figure rose to 72.7%.

The researchers also tested whether the training could help beyond the C and C++ projects CyberForge used. Some models improved on security tasks involving software written in Python, Java and JavaScript.

“We found that the training could carry over to other languages and other cybersecurity tasks,” Suri said.

Manocha said the results point to multiple benefits.

“This approach offers benefits for AI models and cybersecurity,” Manocha said. “It provides higher-quality training signals, scalable generation, broad vulnerability coverage and cross-language generalization. For cybersecurity, it can help produce more capable defense agents and provide coverage beyond known attacks.”

CyberForge addresses the training-data side of defensive AI, but Suri said broader work will also be needed on how security agents are deployed and coordinated across software systems. For defenders, the underlying challenge remains the same: they have to account for more possible weaknesses than an attacker needs to exploit.

“It’s going to be a race,” Suri said. “We should make our agents as well equipped as they can to win.”

—Story by Samuel Malede Zewdu, CS Communications

6 Faculty Members Join UMD’s Department of Computer Science

Samuel Malede Zewdu — Wed, 09 Sep 2026 14:19:47 +0000

New tenure-track and professional-track faculty bring expertise in high-performance computing, internet measurement, computer vision, artificial intelligence, data science and scalable systems.

September 09, 2026

Top row, from left: Yafan Huang, Ritwik Gupta and Ruoshi Liu. Bottom row, from left: Loqman Salamatian, Gheni Abla and Tammy Perrin.

The University of Maryland’s Department of Computer Science is welcoming six faculty members during the 2026-27 academic year, adding research and teaching expertise in high-performance computing, computer networks, computer vision, artificial intelligence, data science and software systems. The appointments include four tenure-track faculty members and two professional-track lecturers joining in fall 2026 and spring 2027.

“We’re thrilled to welcome these new faculty members to the department,” said Department Chair Matthias Zwicker, who holds the Elizabeth Iribe Chair for Innovation and the Phillip H. and Catherine C. Horvitz Professorship. “Their expertise will strengthen research and education across the department, and we are very much looking forward to supporting their work at UMD.”

(UMIACS).

(AIM).

Salamatian’s research focuses on measuring and understanding the internet despite limited visibility into its decentralized infrastructure. His work develops methods to determine what network properties can be inferred from partial observations and establishes theoretical guarantees for those conclusions.

He applies those methods to study how geopolitical events and other disruptions affect internet connectivity and network structure. His research also examines how network measurements can help diagnose performance problems and support more resilient infrastructure.

His work received a Best of CCR recognition from ACM SIGCOMM Computer Communication Review and was selected as a Communications of the ACM Research Highlight.

He earned his doctorate in computer science from Columbia University. Salamatian will join the department as an assistant professor in spring 2027.

Tammy Perrin

Perrin’s teaching and technical interests include artificial intelligence, scalable architectures and high-performance computing. She has taught courses in agentic artificial intelligence, big data systems and data science.

Perrin has more than 15 years of technology industry experience, including work at Google, Capital One and Lockheed Martin.

She earned a master’s degree in computer science from Johns Hopkins University and a bachelor’s degree in physical sciences from the University of Maryland, with studies in electrical engineering, mathematics and physics. Perrin is a two-time recipient of the DCFemTech Powerful Women Programmers Award.

Perrin joined the department as a lecturer in fall 2026 and teaches in the

The annual fall event brought students, faculty and staff together at the Iribe Center to mark the start of the fall semester.

September 08, 2026

Students, faculty and staff gathered at the Brendan Iribe Center for Computer Science and Engineering on August 31 for the University of Maryland Department of Computer Science’s annual Fall Welcome Back Social.

Held on the first day of classes, the social gave new and returning students an opportunity to meet classmates, connect with department representatives and student organizations, and learn about resources available throughout the academic year. The department begins the fall semester with about 3,300 undergraduate students and 612 graduate students.

Previous

Next

—Portrait Article by Samuel Malede Zewdu, CS Communications

NSF Renews Maryland-Led Quantum Simulation Institute’s Funding

Samuel Malede Zewdu — Thu, 03 Sep 2026 14:40:19 +0000

September 03, 2026

A University of Maryland-led institute focused on robust quantum simulation will launch a new phase of research with a five-year U.S. National Science Foundation award expected to total $37.5 million.

The renewal for the which takes effect Sept. 1, reflects UMD’s long-term commitment to advancing quantum computing to take on society’s grand challenges and to spark a new era of scientific, technological and economic development, UMD President a network of interdisciplinary research centers created to accelerate advances in quantum information science through collaborative research, education and workforce development.

“For more than four decades, NSF has been laying the foundational groundwork of research and discovery that is powering today’s modern quantum computing, sensing and communication,” said Andrew Childs, a UMD professor of computer science who directed NSF RQS during its first five years. “I’m grateful for all the collaborations it enabled and proud of the progress RQS researchers made in just five short years.”

Childs is stepping down as director after guiding the institute through its formative years. a physicist at NIST and an adjunct assistant professor in physics and the University of Maryland Institute for Advanced Computer Studies (UMIACS), will serve as deputy director.

Mohammad Hafezi, a Minta Martin Professor of Physics and Electrical and Computer Engineering at the University of Maryland, will become director of the NSF Quantum Leap Challenge Institute for Robust Quantum Simulation under its renewed five-year award. Image courtesy of NSF RQS

“We are at the cusp of determining exactly what quantum computers can and cannot do in the near future,” Hafezi said. “Realizing that potential will require vibrant partnerships among universities, federal laboratories and private industry, and we’re grateful for this renewed support to help lead that effort.”

The new award funds a shift toward what institute leaders call “quantum simulation engineering”—developing quantum technologies that are increasingly scalable and capable of addressing important scientific problems. Researchers will focus on three research themes: interacting fermion simulation, which seeks to model the particles that make up matter; fermion-boson simulation, which examines how matter particles interact with force-carrying particles in complex quantum systems; and dissipative quantum simulation, which studies quantum systems interacting with their environments instead of treating environmental effects solely as unwanted noise.

NSF RQS benefits from technical and administrative support provided by UMIACS, which helps coordinate research activities across the institute’s university and federal partners, Hafezi said.

The institute will also expand its educational mission alongside its research portfolio.

During its first funding cycle, NSF RQS helped launch the University of Maryland’s quantum science and engineering minor, supported K-12 teacher professional development workshops that reached 75 educators, and developed innovative that attracted more than 600 attendees and launched the associate vice president for quantum research and education at UMD, will lead the institute’s education and workforce development efforts moving forward.

“Preparing the future quantum workforce means creating opportunities at every stage—from K-12 classrooms to graduate education and professional careers,” Campbell said. “By connecting students, educators and researchers across that entire pipeline, we can help ensure the talent needed to advance quantum science and technology is ready when the next breakthroughs arrive.”

—Story by UMIACS communications group

UMD Course Connects Computer Science, Math and Sports Analytics

Samuel Malede Zewdu — Tue, 01 Sep 2026 15:16:57 +0000

MATH 440: “Geometry and Data Science of Sports Analytics” gives students experience applying programming, data analysis and machine learning to careers in professional sports.

September 01, 2026

Credit: Adobe Stock Images.

Growing up, .

Lott needed a math elective to fulfill his major requirements, and with his lifelong passion for major league baseball, the class sounded like a perfect fit.

“It didn’t seem like the usual math class,” he recalled. “It was about sports analytics, and I’ve always been interested in athletes’ stats. I felt like I had to take it. And obviously, it’s worked out for me, considering where I ended up after graduation.”

Today, Lott works as a junior software engineer in the Baltimore Orioles' baseball analytics department. He helps maintain the systems that house the club's data and supports an internal website and mobile app used by coaches, scouts, analysts and players to improve their performance. Several types of data, from players’ batting averages to sprint speed, become the foundation for important decisions made by the team’s management, like training optimization or roster development.

“We store data about everything you could imagine—pretty much any stat you can think of surrounding the sport of baseball, including historical records going back many years,” Lott explained. “All those stats are used by analysts trying to gain any type of edge on the field or in the front office.”

Lott credits the experiential math course for helping him navigate his career path toward the Orioles and a professional sports job he always wanted. He said the class, which tied together his background in both computer science and math, allowed him to directly explore complex, real-life scenarios in sports—something that no other college course had ever done.

A math course unlike any other

First piloted in spring 2023, the UMD sports analytics course grew out of Rubinstein’s : “Geometry and Data Science of Sports Analytics.”

For Rubinstein, the new class marks a milestone for the

Five panels introduced the department’s 179 incoming graduate students to research in theory, artificial intelligence, robotics, systems and cybersecurity.

August 31, 2026

The University of Maryland’s Department of Computer Science organized its annual Computer and Computational Sciences Research Day on August 27, giving incoming graduate students an overview of research across computing fields and opportunities to connect with faculty. The event took place in Antonov Auditorium at the doctoral students, 49 master’s students and 52 students pursuing a combined bachelor’s and master’s degree.

Faculty presentations were organized into five panels: “Foundations: Theory, Language, Quantum & Parallelism”; “AI for Science, Society & Real-World Impact”; “Machine Perception & Sensing”; “Robotics, Autonomy & Embodied Intelligence”; and “Systems, Networking & Cybersecurity.” Breaks between sessions gave students time to discuss the presentations with faculty members and one another.

The event featured research from the Department of Computer Science and several UMD institutes, centers and laboratories, including the (MC2), the .

Amol Deshpande, a computer science professor and the department’s associate chair for graduate education, described how the format introduces students to several subjects.

“Research Day gives incoming graduate students an opportunity to learn about work taking place across the department,” Deshpande said. “For those who have not already narrowed their interests, it is a chance to meet faculty members and other students, ask questions, and begin identifying potential advisers or research groups.”

Uzi Vishkin’s presentation introduced him to parallel computing in a simpler way than he had previously examined.

“I would like to not just understand what I want to focus on, but understand every kind of adjacent piece of technology around it,” Mazi said. “Knowing about hardware greatly informs the kind of software you should be writing to take advantage of that.”

The AI panel also connected with Mazi’s interest in health care. He cited Assistant Professor Ritwik Gupta’s discussion of disaster analysis and another presentation on using AI in education as possible connections outside his primary area.

and Ramani Duraiswami on benchmarking and training large audio-language models that process speech, music and background sounds.

Anand said the range of subjects was the main feature that stood out to him during Research Day.

“It was not limited to just one area,” Anand said.

He pointed to presentations on computer vision, robotics, natural language processing, audio processing, dexterous manipulation and diffusion models. Anand said he plans to examine diffusion models, audio understanding and audio processing while determining a specific doctoral research direction.

The department’s graduate program is No. 12 overall and No. 7 among public institutions in

Research co-authored by Gabe Kaptchuk found that combining ad targeting with campaign performance data can expose private user information despite technical safeguards.

August 28, 2026

Tech companies and browser developers have spent years pursuing a seemingly ideal solution to one of the web’s thorniest problems: using advanced cryptography to make online advertising useful without sacrificing user privacy.

New research co-authored at the University of Maryland shows there is a fundamental limit to that goal. Perfect privacy and useful targeted advertising, the researchers found, simply cannot coexist.

The in Canada, challenge a foundational hope in privacy engineering. Rather than treating ad targeting and campaign measurement as separate features, the researchers zoomed out to examine them as two sides of the same coin. In doing so, they revealed that trying to protect user privacy while running both mechanisms simultaneously creates systemic vulnerabilities—a big-picture conflict that previous studies had overlooked by analyzing each piece in isolation.

Gabe Kaptchuk, a study co-author and UMD assistant professor of computer science, said the work demonstrates why the security of an advertising system cannot be evaluated piece by piece.

“Our research shows that when you connect targeting and performance feedback, the interaction introduces a fundamental, unavoidable layer of information leakage,” he said.

Along with Kaptchuk, the research team includes lead author assistant professor of mathematics at Baruch College; the Edwin Sibley Webster Professor of electrical engineering and computer science at MIT.

Kaptchuk, who has an appointment in the University of Maryland Institute for Advanced Computer Studies (UMIACS), said the study grew out of efforts across academia and the technology industry to rethink online advertising amid mounting concern over the extensive tracking that underpins much of the digital ad ecosystem.

“Digital advertising is incredibly invasive, as it relies on gathering a tremendous amount of information about individuals,” he said.

That tracking can cause concrete harms. Commercial ad networks can build profiles around sensitive characteristics—identifying, for example, people believed to be “depression-prone” or “credit-reliant.” Such profiling has been shown to enable predatory marketing, discriminatory job or housing advertising, and manipulative political microtargeting.

But the researchers found that efforts to make advertising more private have overlooked a basic contradiction: The performance information advertisers need to evaluate and refine campaigns can itself reveal information about users.

Modern advertising systems connect two critical functions—deciding who receives an ad and reporting how that ad performs. Even if each function incorporates privacy protections, combining them can create new vulnerabilities.

An advertiser, for example, could repeatedly target carefully selected groups, such as people on a customer list, and analyze the resulting performance metrics. Over time, those results can allow the advertiser to statistically infer private information about users—even when cryptographic protections prevent their identities from being directly revealed.

For browser vendors and ad-tech companies developing the next generation of web advertising standards, the findings suggest that adding privacy protections to individual components will not be enough.

“Simply throwing privacy-enhancing technologies at the problem won't work—we need to be careful about designing interventions,” said Kaptchuk, who is also a core faculty member in the

Computing Catalyst’s TechStart program gave University of Maryland students a week to explore coding, design and technology through community-focused projects.

August 26, 2026

TechStart participants come together for a final group photo after a week of coding, collaboration and project development. Credit: Samuel Malede Zewdu

A tool to help students make connections on campus. A redesigned University of Maryland map. A way to find cooler spaces during hot summer days.

Those were among the projects UMD students presented as the TechStart program concluded. Across separate virtual and in-person sessions, participants spent a week identifying problems, considering potential users and creating technology-based solutions.

TechStart is a paid program for students interested in exploring computing, including those with little or no previous coding experience. This year’s in-person and virtual sessions included students studying computer science, information science, government and politics, and other fields, as well as students still considering a computing major.

Participants took part in programming, design thinking activities and career discussions. They also met with technology professionals who shared perspectives on careers and industry work.

“While TechStart teaches students Python and app development, building community is just as central to the program,” said led sessions on Python data structures, artificial intelligence and prototyping platforms, introducing students to tools and concepts they could use throughout the program.

“TechStart is one of my favorite things to teach,” said Gonzalez, the program’s lead lecturer. “This year, the students did a great job with their industry mentors to solve tech-for-good problems. I’m excited to see how they continue to grow at UMD and with Computing Catalyst.”

The students’ work took different forms as teams moved from early ideas to project prototypes.

Sophomore technology and information design major joined TechStart with no prior coding experience. Her team developed an application concept that would synchronize calendars across educational platforms, bringing deadlines, notifications and academic resources into one place.

“I saw TechStart as an opportunity to bridge the knowledge gap,” Angwi said. “Additionally, I wanted to meet my professors and connect with potential classmates.”

By the end of the program, Angwi said the experience had changed some of her assumptions about working with technology.

“The program has transformed my approach to tech,” Angwi said. “It has taught me that it demands far more than mere code copying.”

Program staff members saw similar changes as students moved through the week.

A UMD-led team, including Chris Metzler, Abhinav Bhatele and Haizhao Yang, is developing an AI framework to produce faster, more detailed long-range water forecasts.

August 24, 2026

Predicting whether next season will bring a devastating flood or a crippling drought often comes down to a race against the clock. For climate scientists, forecasting the future movement of water requires some of the most complex—and computationally demanding—math on Earth.

Across the United States, water-related hazards threaten electrical grids, reservoirs and other critical infrastructure. Energy providers need localized, long-range forecasts to prepare for these risks, but current prediction systems are limited by scientific uncertainty and the enormous computing power required to run them.

Now, a University of Maryland-led team is using generative artificial intelligence to change that.

The researchers have been awarded $750,000 in Phase I funding from the U.S. Department of Energy's new Genesis Mission to develop an AI-powered framework that can generate faster, more detailed seasonal and multi-year water forecasts. The nine-month award includes approximately $470,000 for UMD, the lead institution.

The project, "An Agentic AI Framework for Seasonal-to-Interannual U.S. Water Prediction," was announced last month at the inaugural Genesis Mission Summit in Washington, D.C. It was one of 270 projects selected from more than 5,000 proposals submitted to the highly competitive national program. If chosen for Phase II, the team will expand its early prototypes into operational tools with broader real-world applications.

Led by Jinwoong Yoo, an assistant research scientist at UMD's : Maria Molina, assistant professor of atmospheric and oceanic science; Chris Metzler, assistant professor of computer science; Abhinav Bhatele, professor of computer science; and Haizhao Yang, professor of mathematics and Affiliate Professor of computer science.

Researchers from Oak Ridge National Laboratory, the University of Florida, the University of Colorado Boulder and Silurian AI are also collaborating on the effort.

For Molina, the partnership reflects a growing need to combine advances in AI with decades of expertise in Earth system science. The team's goal is to integrate cutting-edge AI forecasting with the Department of Energy's flagship Energy Exascale Earth System Model (E3SM), creating a system that can produce faster predictions without sacrificing scientific rigor.

“If we can get this framework right, we can give planners a clearer picture of the hazards their communities may face seasons to years in advance by revealing how event likelihood, intensity and local impacts may evolve,” Molina said.

Traditional physics-based Earth system models such as E3SM simulate interactions among the atmosphere, ocean, land and water cycle across the globe. Seasonal-to-multiyear forecasting requires many simulations to capture the range of possible outcomes, creating tradeoffs among resolution, forecast length, and ensemble size. Turning those forecasts into useful information at the community level remains difficult because predictability decreases at longer timescales, while local hazards depend on processes that global models cannot fully represent.

The UMD team's approach uses AI to accelerate nearly every step of that process.

The framework will first incorporate satellite observations of soil moisture and groundwater using machine learning to improve the model's starting conditions. Generative AI models trained on decades of E3SM simulations will then emulate the behavior of the original model, producing more than 100 possible forecast scenarios to estimate both likely outcomes and uncertainty more than 100 times faster than traditional methods.

Image-enhancement techniques will sharpen those regional forecasts into street-level flood maps with one-meter resolution. Behind the scenes, autonomous AI software agents will coordinate the complex workflow—from preparing model inputs to running simulations and analyzing results—allowing researchers to focus on evaluating forecast quality and scientific reliability.

To ensure the predictions remain grounded in physical science, Molina will lead the team's validation efforts. Researchers will test the framework against major hurricanes, evaluate whether it captures climate patterns such as El Niño and use explainable AI techniques to determine which factors drive the model's forecasts.

"The real bottleneck has always been computational cost," Metzler said. "By using AI to emulate these massive Earth system models, we can run simulations in minutes on supercomputers that used to take days. That speed is what makes real-time protection of local grids actually possible."

Before deploying the system on the Department of Energy's leadership-class supercomputers, the team will develop, train and evaluate its initial models using UMIACS' high-performance computing infrastructure.

Ultimately, the researchers hope their work will help utilities, water managers and emergency planners make better decisions before disasters strike.

For the team, success won't be measured simply by faster algorithms or more efficient computing. It will be measured by more accurate forecasts that help communities prepare for floods, protect critical infrastructure and better manage water resources before extreme conditions arrive.

—Story by Melissa Brachfeld, UMIACS communications group

The Code They Couldn't Talk About

Samuel Malede Zewdu — Fri, 21 Aug 2026 19:07:10 +0000

UMD alum Eileen Buckholtz spent three decades working with emerging technology at the NSA. Decades later, she is documenting the computing careers of women whose classified work kept them largely out of history.

August 21, 2026

For much of the career of alum ,” a project documenting the women who worked in computing at the NSA.

“There were more than 150 of us, but I wrote about 25 in my book,” Buckholtz said. “We were almost all hired as 21- and 22-year-old women, and we began to program and manage the most sophisticated computers in the world.”

Early computing

Buckholtz entered the field when computer science was still developing as an academic discipline. After earning an undergraduate degree in computer science from Ohio State University, she joined the NSA at 21.

She later earned her master’s degree in computer science from the University of Maryland in 1973, the same year UMD established its Department of Computer Science.

At Maryland, Buckholtz encountered technologies and ideas that were reshaping computing. A discussion in a data communications course introduced her to ARPANET, the early network that preceded the modern internet. At NSA, she connected to the network for the first time and sent a message to a researcher at the Stanford Research Institute, now SRI International, in California.

“There were a lot of possibilities,” she said.

Her master’s research explored potential computer applications in a cable television environment, including concepts such as online education and shopping. Her Victor Basili, was later published in Communications of the ACM, and she presented her research at the 1973 ACM Annual Conference.

Buckholtz also carried lessons from UMD back to the NSA. After taking a course on formal requirements for projects, she brought the methodology to the agency, where she said it was later adopted for requirements specification.

A Career Behind Secrecy

During her 31-years at the NSA, Buckholtz worked across a range of emerging technologies and computer systems.

She served as the first project manager for a multimillion-dollar on Nov. 17, 2026.

Other UMD graduates featured in the book include:

Carol Osborne McWilliams (B.A. ’67, english)

Peggy Ann Grindrod Strader (B.A. ’66, education)

Kathy Reading (B.S. ’98, management)

Mary T. Clulow (B.S. ’97, information systems management)

Readers, including students interested in learning more about Buckholtz’s work, can connect with her through the Queens of Code .

CS Researchers Receive AIM Seed Grant to Study the Roots of AI Bias

Samuel Malede Zewdu — Tue, 18 Aug 2026 19:13:24 +0000

Naomi Feldman, Rachel Rudinger and Sarah Wiegreffe will examine how cognitive biases influence language model decisions and test methods for reducing them.

August 18, 2026

A team of University of Maryland Department of Computer Science researchers has received seed funding for an interdisciplinary effort that combines cognitive psychology and computer science to investigate how artificial intelligence makes decisions—and why those decisions can become irrational or biased.

Supported by a one-year grant totaling just under $120,000—half from the .

The collaborators are Naomi Feldman, professor of linguistics; Rachel Rudinger, associate professor of computer science; and Sarah Wiegreffe, assistant professor of computer science. All three are members of UMD’s Computational Linguistics and Information Processing (CLIP) Lab, which is supported by UMIACS.

The CLIP researchers are tackling a fundamental challenge in modern AI. Although large language models are increasingly used to support decisions in areas such as hiring and college admissions, they can be influenced by information that should have no bearing on the outcome. In human psychology, these errors are known as cognitive biases—for example, the anchoring effect, in which an unrelated number unconsciously influences a person's estimate or judgment. The UMD team has found that language models exhibit many of these same irrational tendencies.

Feldman's lab brings extensive expertise in computational cognitive science—using computer models to understand human thought processes—and is now reversing that approach by applying theories of human cognition to better understand machine behavior.

“We’ve known for a long time that people reason irrationally,” said Feldman. “Anytime people need to estimate a quantity or proportion to decide, they are subject to cognitive biases. In any decision-making context involving these models, we want to make sure decisions are based on facts in a rational manner rather than on irrelevant information in the surrounding context.”

While these cognitive biases may seem relatively harmless, the researchers believe they could also create pathways for harmful social biases, such as racial or gender stereotyping, to influence AI-assisted evaluations like resume screening and admissions decisions.

“Cognitive bias and social bias are often studied separately, but we think they're deeply connected,” said Rudinger. “If we can understand why language models rely on irrelevant information when making decisions, we'll be in a much better position to reduce the unfair biases that can emerge in real-world applications.”

To move from observation to intervention, Wiegreffe’s lab focuses on mechanistic interpretability—the study of the internal circuits that drive a model's behavior. By identifying the specific components that respond to irrelevant context or demographic attributes, the team hopes to eventually disable, or “ablate,” those circuits without diminishing the model's overall capabilities.

“The key challenge is that models are undesirably influenced by context, but this influence is inconsistent,” said Wiegreffe. “When AI systems are used in high-stakes settings like job interviews and resume reviews, our goal is to give people tools to control those models. We want users to be able to ablate specific circuit components so they can trust that the model is truly ignoring race or gender in a predictive decision.”

Feldman said the collaboration grew out of work by CLIP graduate students. Her Ph.D. student, Hillary Owusu, began investigating AI cognitive biases for a class project and discovered strong connections with research on AI and human reasoning led by Rudinger and her Ph.D. student, Shramay Palta. The AIM grant will provide full support for both senior doctoral students over the coming year.

The work also relies heavily on UMIACS’ advanced computing infrastructure. Because analyzing the internal representations of large language models requires significant computational power, the researchers use the CLIP and Over the next year, the team plans to expand its dataset of cognitive testing benchmarks, evaluate new intervention strategies and use the results to pursue a larger collaborative proposal to the National Science Foundation. Ultimately, the researchers hope their work will not only explain why AI systems make biased decisions but also provide practical tools that help people identify and reduce those biases before AI is used to make consequential decisions.

—Story by Diya Sharma, UMIACS communications group

Noriko Maitland Receives UMD Provost’s Academic Advisor of the Year Award

Samuel Malede Zewdu — Tue, 18 Aug 2026 14:30:40 +0000

She was selected for the 2025-26 academic year for her work on advising resources, student communication and departmental initiatives.

August 18, 2026

University of Maryland Department of Computer Science Academic Advisor Noriko Maitland has received the Provost’s Academic Advisor of the Year Award for the 2025-26 academic year.

Each year, a committee of academic advisors selects three university employees for the award, which recognizes advising contributions across campus. Maitland will receive a plaque and a $1,000 award and will be recognized at the Undergraduate Studies Annual Academic Advising Conference on October 7.

“I received the award notification email on my birthday, so it was a big surprise to me,” Maitland said. “I thank my team in CS Undergraduate Advising, especially Madeline G., for encouraging me and supporting my ideas for the projects.”

According to her nomination materials, Maitland contributed to several initiatives to address student advising needs and improve how the department delivers information and resources.

Among those efforts was a collaboration with UMD’s International Student & Scholar Services to develop additional student support. She also helped design and launch the department’s advising ELMS platform, which brings advising information and resources together in one location.

The nomination noted that Maitland incorporated student feedback into platform updates as the advising team continued developing the resource.

Her work also included changes to the department’s advising newsletter. Maitland developed processes for organizing and distributing the newsletter to provide students with academic, career and advising information in a more consistent format.

For Maitland, interactions with students have played a central role in shaping those efforts.

“It has been great, and I enjoy working with students in the CS program,” Maitland said of her advising experience at UMD. “They are my inspiration to make the advising service better.”

Her selection also reflects work carried out collaboratively within the department’s undergraduate advising office.

“It’s nice to be recognized, but I truly believe the award is for the entire CS Advising team,” she said.

—Story by Samuel Malede Zewdu, CS Communications

CMMRS Marks 10 Years of Connecting Students With CS Research

Samuel Malede Zewdu — Fri, 14 Aug 2026 14:19:12 +0000

The weeklong research school brings together students and faculty from UMD, Cornell and the Max Planck Institutes to explore computer science research and graduate study.

August 14, 2026

The and the Ramani Duraiswami participated in this year’s program, delivering a David Guan, who participated in this year’s program, said the speakers made their research accessible even to students who entered the sessions without prior knowledge of their fields.

“I was especially surprised by how open and welcoming the faculty were throughout the program,” Guan said. “They often joined us for meals and were always happy to discuss their work or offer advice.”

Guan said another part of the experience was meeting students from different backgrounds and learning about their perspectives.

“Being part of CMMRS was exciting and felt like a rare opportunity to be surrounded by talented students from around the world,” Guan said. “I really enjoyed hearing how their cultures and experiences differed.”

That international exchange was also reflected in the composition of the 2026 cohort, which included students from India, Nigeria, Rwanda, Germany, Bulgaria and the United States.

For participants considering doctoral study, the program also provides opportunities to hear directly from graduate students about their experiences. Connor Clayton, a UMD computer science Ph.D. student and 2026 CMMRS mentor, said the school helps students understand what pursuing a doctorate can involve.

“At CMMRS, students considering a Ph.D. can expect to receive a lot of relevant guidance, including insight about how to decide if a Ph.D. is right for them, what grad student life is like and how to apply,” Clayton said.

—Story by Samuel Malede Zewdu, CS Communications

###

For students interested in learning more about CMMRS, including future program dates, eligibility and application information, visit the

Selected from nearly 400 applicants, the students spent 10 weeks working across seven research teams in computer science and mathematics.

August 11, 2026

27 undergraduate students concluded 10 weeks of research at the University of Maryland’s Department of Computer Science on Aug. 7, completing projects that ranged from artificial intelligence evaluation to mathematical theory as part of the department’s summer Research Experiences for Undergraduates (REU) Combinatorics, Algorithms, and AI for Real Problems (CAAR) program. Selected from nearly 400 applicants, this year’s cohort worked alongside UMD faculty and graduate student mentors across seven research teams.

The REU-CAAR is an William Gasarch, a professor of computer science who has helped organize the program since 2013, said students increasingly arrive with prior exposure to research and some familiarity with the subjects they plan to study.

“The cohorts get better and better every year in that students are starting research earlier and earlier,” Gasarch said. “Many of them started doing research in high school, which would have been unusual the first year I ran the program.”

That preparation helped students move from studying existing work early in the summer to investigating questions of their own as the program progressed. The process differed by project, with some teams working on theoretical problems and others examining computational systems and applications.

One group studied Ramsey theory, an area of combinatorics that examines when patterns must appear within sufficiently large mathematical structures. The project led by Gasarch initially focused on understanding established results and developing clearer ways to explain them, including material that could eventually be presented to high school students.

As the group worked through those problems, the project moved beyond its original educational focus and led to new mathematical findings.

’s experience. Truong, a junior double majoring in computer science and mathematics, worked on a project led by Professor Jordan Boyd-Graber, examining how artificial intelligence systems respond to challenging questions that combine visual and language information.

The research looks beyond whether an AI system produces a correct or incorrect answer. Instead, the team is studying where systems perform well, where they encounter difficulty and how their capabilities compare with human performance.

Truong said more detailed evaluation methods could give researchers a clearer understanding of what current AI systems can and cannot do.

“More meaningful and fair evaluations can help researchers make more honest claims about AI progress and identify areas that still need improvement,” Truong said. “Ultimately, this work could contribute to building AI systems that are more reliable and work more effectively alongside people.”

Carrying out that research also required students to coordinate their individual responsibilities with the goals of a larger team. Truong’s group met weekly to discuss progress, work through challenges and determine the next steps for the project.

“Working with the group taught me the importance of being organized, communicating clearly and connecting my individual work to a shared research goal,” Truong said.

He also pointed to interactions with students from other institutions as an important part of the summer. Participants entered the program with different academic experiences, giving them opportunities to compare approaches to research and problem-solving.

“Hearing their perspectives and seeing how they approached research broadened the way I think about solving problems,” Truong said.

Across the seven teams, participants learned how research develops from early reading and discussion into more focused investigation. Seminars supplemented that work by introducing participants to faculty research and providing opportunities to discuss graduate study and academic careers.

Gasarch said the program is also intended to expose students to a defining feature of research: unlike coursework, the problem being studied may not have a predetermined solution and can change as new information emerges.

“They will learn how to work in a group and learn that when doing research, the answers are not known, and sometimes the questions change,” Gasarch said.

Other 2026 REU projects:

• Laxman Dhulipala led a project on I/O-efficient parallel algorithms. Students explored theoretical and practical approaches for designing parallel algorithms that reduce the amount of data transferred between main memory and external storage while maintaining high levels of parallelism. The project included developing data structures and algorithms for working with large amounts of data distributed across external memory devices such as solid-state drives.

• Erin Molloy led a computational biology project focused on errors in phylogeny reconstruction. Students examined how errors introduced during genomic data processing can affect algorithms used to reconstruct evolutionary relationships among species or cells. The project included studying error propagation, developing algorithms that are more robust to uncertain inputs and exploring machine learning methods for detecting errors.

• Bill Regli and Bill Gasarch led a project exploring alternative approaches to computing beyond traditional transistor-based systems. Students investigated technologies including memristors, phononic systems, Ising machines and analog computers to understand how they could address computationally difficult problems. The project examined the potential advantages and limitations of these approaches for problems such as Boolean satisfiability, fast Fourier transforms and simulations of physical systems.

• Ming Lin led a machine learning project focused on autonomous driving. Students examined how machine learning systems can use human driving data to better anticipate behavior in environments where autonomous and human-driven vehicles operate together. The project focused on high-risk driving scenarios and applying machine learning techniques to improve understanding of driving dynamics and vehicle safety.

• Sarah Miller led a project exploring alternative neural network architectures for solving games, puzzles and real-world reasoning problems. Students used ideas from combinatorial game theory to build and train neural networks designed to reason efficiently in unfamiliar situations. Depending on their interests, students could extend the work to physical systems, formal verification using Lean or quantum computing while comparing the approaches with transformer-based and reinforcement learning methods.

—Story by Samuel Malede Zewdu, CS Communications

TRAILS Awards More Than $515K for Trustworthy AI Research

Samuel Malede Zewdu — Mon, 10 Aug 2026 16:17:58 +0000

The institute, directed by Hal Daumé III, selected five multidisciplinary projects focused on AI literacy, autonomous drones, public trust and local government policy.

August 10, 2026

Helping students learn to evaluate AI. Making autonomous drones safer. Measuring how people express trust in AI. Helping local governments navigate AI policy. Understanding how to reestablish trust after AI failures.

Those are among the questions driving five new research projects selected for the fourth round of seed funding from the Hal Daumé III, director of TRAILS and a professor of computer science at UMD. “Over the past few years, we've moved the conversation beyond model accuracy alone to examine how AI systems perform in real-world human, institutional and societal contexts.”

The five projects illustrate how that vision can be put into practice.

“We're continuing to transition from cutting-edge research and scholarship to research with significant impact,” said Rebuilding Trust After AI Failures

Broniatowski and Cornell’s the team will examine how people’s values shape their perceptions of AI risks, responsibility and accountability, and whether meaningful explanations tailored to different worldviews are more effective than detailed technical information alone at rebuilding public trust.

—Story by UMIACS

CS Faculty Join Federal Initiative to Speed AI-Powered Discovery

Samuel Malede Zewdu — Fri, 07 Aug 2026 17:08:19 +0000

August 07, 2026

Four faculty members from the University of Maryland Department of Computer Science, who are affiliated with the a new initiative designed to accelerate AI-driven scientific discovery.

Chris Metzler, Abhinav Bhatele, and Haizhao Yang, who will help develop an agentic AI framework to overcome computational bottlenecks in the DOE's long-term climate forecasting system. The team aims to improve predictions of floods and droughts, leading to more accurate, localized and long-range forecasts that can help protect critical infrastructure such as power grids and water supplies from extreme weather.

The project, led by Jinwoong Yoo, an assistant research scientist at the University of Maryland and NASA Goddard Space Flight Center, received $750,000 in Phase I funding.

UMD computer scientist Xiaodi Wu is part of a Virginia Tech-led project that will use AI to design more efficient quantum error-correcting codes, helping improve the accuracy and resilience of quantum information. Current approaches require extensive computing resources to protect even a single unit of quantum information, and designing better methods is computationally prohibitive. By using AI to accelerate the design process, the team hopes to develop tools that will speed advances across the quantum computing research community. The project received $720,000 in Phase I funding.

Awardees will gain access to the Genesis Mission Platform, which includes AI agent frameworks, advanced AI models and software from industry partners, and high-performance computing resources at DOE national laboratories and partner facilities. These resources will allow researchers to rapidly design, test and refine new approaches to scientific discovery. Teams selected for Phase II will scale their initial prototypes into tools with broader real-world impact.

—Story adapted from a news brief published by UMIACS

Learning When to Pivot

Samuel Malede Zewdu — Fri, 07 Aug 2026 14:02:48 +0000

CS alum Ananth Sankaralingam turned early startup experiments into a Y Combinator-backed AI platform for government contractors.

August 07, 2026

Two weeks into an internship at Amazon, alum and the Furong Huang gave him a closer look at large language models and the range of products that could be built around them.

“You realize there’s so much more possible at this moment,” Sankaralingam said. “The opportunity to be at the forefront of that, when the world is changing so drastically, is a once-in-a-lifetime opportunity that I didn’t want to pass up.”

Changing Course

Sankaralingam later co-founded Chromie.dev with

CS Ph.D. student Yvonne Zhou developed a framework that protects sensitive data during and after model training while reducing training time by up to 15-fold.

August 05, 2026

As artificial intelligence becomes increasingly integrated into sectors such as healthcare, finance and government, organizations face a critical challenge: how to use powerful machine learning tools without compromising confidential data.

At the University of Maryland, fifth-year computer science doctoral student Yvonne Zhou is developing a new framework designed to bridge the gap between privacy-preserving theory and practical, real-world performance. Her work focuses on protecting sensitive information throughout the machine learning pipeline—from the moment data is sent to a cloud server through deployment of the final model.

“Many organizations don’t have the computational resources to train large models themselves and therefore need to outsource training to cloud providers,” Zhou said. “The challenge is how to take advantage of external computing power without exposing private data.”

Zhou’s research addresses what experts consider a major bottleneck in privacy-preserving AI. Existing cryptographic techniques such as fully homomorphic encryption (FHE) allow cloud servers to train models using encrypted data, enabling computation without revealing the underlying information. But the process is notoriously slow and computationally expensive, and it protects data only during training. Once a model is decrypted and deployed, it can still leak sensitive information.

To address this limitation, Zhou developed a framework that combines the rigorous theoretical guarantees of FHE with a customized differential privacy algorithm, which introduces carefully calibrated random noise to help prevent sensitive information from being exposed.

“Together, FHE protects data during outsourced training, while differential privacy protects against information leakage after deployment,” Zhou explained. “In other words, our framework provides end-to-end privacy guarantees.”

A key breakthrough was overcoming the enormous computational cost of combining these two techniques. By developing a new training algorithm that eliminates a particularly expensive computational step, Zhou’s framework achieved up to a 15-fold reduction in training time while maintaining accuracy comparable to significantly slower approaches.

The findings are described in a in South Korea. The work is part of an ongoing collaboration supported by Dana Dachman-Soled, an associate professor of electrical and computer engineering and a core member of the Min Wu, a professor of electrical and computer engineering. Both hold appointments in the University of Maryland Institute for Advanced Computer Studies (UMIACS).

Dachman-Soled said Zhou quickly mastered entirely new areas while working on the project. Despite having no previous background in fully homomorphic encryption or C++, she was implementing complex encrypted training pipelines within months.

"Beyond her technical abilities, she has consistently demonstrated exceptional initiative and persistence, making her an outstanding researcher and an invaluable member of the team," Dachman-Soled said.

Wu said Zhou's work tackles one of the central challenges in trustworthy AI: balancing strong privacy guarantees with practical performance.

"By combining rigorous theory with algorithmic innovation, she has developed an approach that makes secure machine learning substantially more practical while maintaining strong privacy protections," Wu said.

While the current paper focuses on foundational models that establish rigorous mathematical guarantees, Zhou sees it as an important first step. The team’s next challenge is extending these methods to the deep neural networks that power today’s most advanced AI systems.

“I’m excited to continue exploring how we can make privacy-preserving machine learning both theoretically well-founded and practically useful for the large-scale AI systems that people rely on every day,” Zhou said.

—Story by Diya Sharma, UMIACS communications group

Making Science Easier to See

Samuel Malede Zewdu — Tue, 04 Aug 2026 13:47:52 +0000

Zhicheng Liu and Fumeng Yang received a $559,048 NSF grant to develop tools and design guidelines for more effective science infographics.

August 04, 2026

Scientific discoveries can influence decisions about health, climate and technology, but the research behind them is often difficult to explain outside laboratories and academic journals. Two University of Maryland computer scientists are examining how infographics can present those findings to the general public effectively.

Department of Computer Science Associate Professor Zhicheng Liu and Assistant Professor Fumeng Yang received a $559,048 (NSF) to study how science infographics are designed, interpreted and evaluated.

The researchers will study how professional designers choose layouts, images, labels and other visual elements to explain scientific information. They will also examine how those decisions affect what audiences understand, remember and consider credible.

“Our goal is to make it easier for scientists, journalists and designers to communicate scientific discoveries and knowledge through effective and trustworthy infographics,” Liu said.

Infographics are used in news stories, scientific publications and classrooms to explain subjects such as public health, climate change and space exploration. Unlike a conventional chart, an infographic may combine data, diagrams, illustrations, labels and narrative elements in a single visual presentation.

That flexibility can make scientific information more accessible, but it also creates design challenges. Existing software often allows users to arrange text and images without offering guidance on how the elements should work together or whether the finished product communicates its intended message.

“There are few tools or guidelines that help creators design them effectively,” Liu said. “Our project will develop the scientific foundations, software tools and evidence-based design guidelines needed to support the creation of effective science infographics.”

Liu will lead the development of what the researchers call a “grammar” of infographics. Similar to how a language grammar explains how words and sentences fit together, the framework will describe the components of an infographic and the relationships among data, visual elements and storytelling techniques.

The framework will serve as the basis for a JavaScript-based authoring toolkit. The open-source software is intended to help users connect scientific information with visual components and communication goals rather than treating each chart, illustration or annotation as a separate design element.

Yang will focus on how people create and respond to infographics. Her research will examine how designers make decisions, including which information to emphasize, how to organize a visual explanation and what techniques to use to guide a reader through it.

She will also develop a framework for measuring whether an infographic achieves its intended purpose. The evaluation process will consider factors such as what viewers understand, what information they retain and whether they consider the presentation credible.

The project will also produce datasets that other researchers, educators and software developers can use. Yang feels making those resources publicly available could allow others to test the findings, improve the tools and apply them in different areas of science communication.

“Ultimately, we hope to lower the barriers to effective science communication,” Yang said, “helping more people understand complex scientific ideas and make informed decisions.”

—Story by Samuel Malede Zewdu, CS Communications

Making Complex Math as Easy as Asking a Question

Samuel Malede Zewdu — Wed, 29 Jul 2026 18:58:40 +0000

Haizhao Yang is developing an AI platform that turns everyday questions into mathematical models for complex decision-making.

July 29, 2026

When an emergency like extreme weather strikes, making the right decision in real time can become a matter of life and death. Yet the mathematical optimization tools used to calculate the safest evacuation routes or allocate emergency resources typically require significant computing power and specialized expertise in programming and mathematics.

Haizhao Yang, an associate professor with joint appointments in the University of Maryland's Department of Mathematics and Department of Computer Science, wants to change that. Supported by a and is also a core member of the

Mihai Pop and Philip Resnik delivered keynote addresses and held leadership roles at international conferences focused on artificial intelligence and computational biology.

July 27, 2026

University of Maryland computer science faculty members Mihai Pop and Philip Resnik joined fellow UMIACS faculty member Joshua Weitz in taking on leadership and speaking roles at major international conferences in July 2026. Their participation highlighted research across artificial intelligence, computational biology and microbiome science.

Philip Resnik, a professor of linguistics and an affiliate professor of computer science, delivered the opening keynote address at the , argued that the commercial success of AI and large language models (LLMs) has disrupted the balance of priorities in the computational linguistics community. He outlined three key challenges facing the field and encouraged researchers to approach them through constructive dialogue rather than opposition.

“I wanted to bring conversations already happening across the field into the open and frame them in a way that encourages productive discussion,” Resnik said. “This community played a pivotal role in creating the AI revolution and I argued that we’ve got a crucial and distinctive role to play in how AI evolves moving forward.”

Mihai Pop, an MPower Professor of Computer Science, served as co-chair of the 2026 Conference on a companion event that introduces middle and high school students to computational biology and bioinformatics.

In his keynote, My Long and Winding Road to Computational Microbiome Analytics, Pop reflected on the unexpected path that led him to computational microbiome research, sharing lessons from both successes and setbacks before offering advice to students beginning careers in science.

Joshua Weitz, a professor of biology and the Clark Leadership Chair in Data Analytics, co-chaired the Bioinformatics and AI sessions at the

The University Career Center survey shows strong employment outcomes, higher starting salaries and continued graduate school enrollment for the class of 2025.

July 22, 2026

The University of Maryland’s Department of Computer Science reported a 95% job placement rate for its class of 2025, according to a survey conducted by the Matthias Zwicker, who holds the Elizabeth Iribe Chair for Innovation and the Phillip H. and Catherine C. Horvitz Professorship. “It's rewarding to see the strong career outcomes our graduates continue to achieve.”

The survey tracked outcomes for 1,188 graduates and collected data from 906 students, resulting in a 76% knowledge rate. Among respondents, 78.5% entered the workforce and 16.2% continued their education. Only 1.5% of graduates were reported as unplaced at the time of the survey, while 3.8% had not yet finalized their post-graduation plans.

Graduates also reported higher starting salaries than the previous year. The median starting salary increased to $119,000, with the middle half of respondents earning between $90,000 and $145,000. More than 200 graduates also reported receiving employment bonuses, with a median bonus of just over $10,300.

Internships continued to be one of the strongest predictors of employment. More than 85% of respondents completed at least one internship during their time at Maryland, and nearly 60% completed multiple internships. Among graduates who described how they found their jobs, internships were the most frequently cited pathway, ahead of online job sites, personal contacts and career fairs.

Graduates accepted positions with employers across the technology, finance, government and defense sectors. Companies hiring multiple UMD graduates included Amazon, Capital One, Meta, Bloomberg, Google and Microsoft, while others joined startups, federal agencies and research organizations. Software engineering remained the most common career path, with graduates also entering roles in artificial intelligence, cybersecurity and data science.

Although many graduates accepted jobs across the country, nearly 45% remained in Maryland, Virginia or Washington, D.C. Others relocated to technology hubs including California, New York and Washington state.

Graduate school also remained a popular option. Of the 146 graduates who continued their education, nearly nine in 10 enrolled in master's programs, most commonly in computer science. More than half chose to remain at UMD for graduate study.

“Our students define success in many different ways,” Zwicker said. “Whether they enter the workforce, continue their education or pursue entrepreneurial ventures, we're pleased to see them building on the experiences they had at UMD.”

—Story by Samuel Malede Zewdu, CS Communications

Three CS Terps Received CMNS Alumni Network Endowed Undergraduate Awards

Samuel Malede Zewdu — Tue, 21 Jul 2026 18:11:32 +0000

Raghav Thind, Nghia Truong and Priya Tyagi received funding to support summer research, conference travel and career development.

July 21, 2026

Every year, the College of Computer, Mathematical, and Natural Sciences (CMNS) Alumni Network offers awards to help undergraduates defray the costs of conducting research, attending conferences or interning over the summer.

Read below how this year’s award recipients are furthering their career development with funding from the

She discusses her path into computing, her work on privacy and data de-identification, and why communication matters in human-centered security research.

July 20, 2026

University of Maryland Professor of Computer Science Michelle Mazurek studies how people navigate security and privacy decisions in everyday digital systems. Affiliated with the (MC2), her research connects technical questions with the human choices behind them, from how software developers handle secure coding to how researchers protect sensitive data before sharing it.

In this Q&A, Mazurek discusses the early experiences that drew her to computing, her work on data de-identification and why communication is central to human-centered security research.

Was there a defining moment that shaped your path into computer science?

There was more than one. Growing up, my dad was an electrical engineer, and he taught my sister and me that engineering was a way to solve problems, whether something broke in the house or needed to be fixed on a computer.

I first tried programming in second or third grade with Logo, where you could move a pointer on a screen by writing commands. Later, we used Lego Logo, an early version of Lego robotics systems. We built things with Legos and programmed motors to make them move. That was probably how I became interested in programming and computer science.

Can you describe your research focus?

My work is at the intersection of security and privacy with human-computer interaction. We study how people make decisions related to privacy and security and how to help those decisions go better.

That includes end users trying to protect themselves, software developers whose code affects others and professionals such as librarians who manage privacy issues for the people they serve.

What is one current project you are working on?

One project I am working on with my student

UMD Ph.D. student Ryan Synk is developing a machine-learning system designed to help researchers search a growing federal archive of raw genetic information.

July 16, 2026

Ryan Synk, a doctoral student in computer science at the University of Maryland.

A single biological sample can produce millions of fragments of genetic information. Multiply that output by years of experiments involving humans, animals, plants, viruses and environmental samples, and researchers face a problem that is as computational as it is biological: How do they find the DNA sequence they need?

University of Maryland computer science Ph.D. student Ryan Synk is studying whether machine learning can make that search faster and more reliable.

Advised by Professor of Computer Science Ramani Duraiswami, Synk is developing , or SRA. The federal repository contains raw sequencing data submitted by researchers from around the world, making it a resource for studying organisms, diseases and genetic variation.

Its size and structure, however, also make it difficult to navigate.

“The very thing that makes it really useful to researchers is also the thing that makes it hard to analyze well,” Synk said. “It’s just really big.”

Math Meets Computing

Synk, who earned an undergraduate degree in mathematics and computer science from UMD in 2020, did not initially plan to pursue computer science research. He began as a mathematics major and added computer science after taking courses required by the math program.

“I found it to be a fun way of applying this mathematical thinking that I had to real problems,” Synk said.

His interest in machine learning was influenced in part by AlphaGo, the artificial intelligence system that defeated a professional Go player. As Synk took more computer science courses, he found that he preferred building programs to being tested on computer science concepts.

“I really hated taking CS exams, but I really loved doing CS projects,” Synk said. “I found myself constantly wanting to work on coding stuff.”

A numerical analysis course helped connect his interests in mathematics, science and programming. The class examined how scientific questions, including climate simulations, can be translated into algorithms that computers can run and evaluate.

Synk later worked with UMD Professor of Computer Science Tom Goldstein through the Research Experience for Undergraduates program (REU) and completed an internship at Lawrence Berkeley National Laboratory. Those experiences introduced him to scientific computing, large computing clusters and academic research.

Searching Raw Sequences

Synk’s doctoral research now combines machine learning and scientific computing. LOCALE grew out of his work on vector search, a method that represents information as lists of numbers called vectors. This enables computers to identify similar items by the similarity of their vectors.

Searching the SRA differs from searching a traditional DNA reference database. Sequencing machines typically break DNA into many short fragments called reads. Researchers receive those pieces rather than one complete, neatly ordered genome.

“It’s all there in this more raw, unprocessed state, and that’s what makes it hard to work with,” Synk said.

July 15, 2026

An AI model may be designed to behave safely, but that doesn’t mean the system delivering it is secure, according to new research from the University of Maryland.

The Michelle Mazurek, a professor of computer science with an appointment in the “Our research shows that the software infrastructure managing and scheduling those requests introduces a completely separate, highly complex layer of vulnerability.”

The study was led by Yunze Zhao, a second-year doctoral student in computer science advised by Mazurek. The research team also includes Alan Liu, assistant professor of computer science with an appointment in UMIACS and a core member of MC2, and Yibo Zhao, a second-year doctoral student in computer science. They collaborated with

He was recognized for research on energy-efficient sensing systems that use sound, radio and light.

July 13, 2026

University of Maryland Department of Computer Science alum for developing sensing technologies that operate within limited power, size and computing resources.

Garg, who was advised by Associate Professor of Computer Science Nirupam Roy, is now an assistant professor of electrical and computer engineering at Rice University. He received the award for his dissertation, “

Twenty-seven undergraduates are working with UMD researchers on projects in computational biology, artificial intelligence and parallel computing through the NSF-funded REU-CAAR program.

July 13, 2026

Computer science and mathematics can seem abstract in the classroom. But at the University of Maryland this summer, 27 undergraduates are applying those concepts to problems ranging from reconstructing evolutionary trees to probing the limits of artificial intelligence.

The students are participating in the Research Experience for Undergraduates in Combinatorics and Algorithms for Applied Research (REU-CAAR), a competitive, 10-week National Science Foundation-funded program. Selected from nearly 400 applicants, this year's cohort is working alongside UMD faculty and graduate student mentors on seven research teams tackling a broad range of computational and mathematical challenges.

The program concludes Aug. 7 with students presenting the results of their summer projects. Participants also attend weekly seminars, professional development workshops and campus events that introduce them to graduate-level research.

“The goal is to give students an on ramp into research,” said William Gasarch, professor of computer science and the program's principal investigator. "They're learning not only the technical skills, but also how research works as a collaborative process."

Several research teams are led by faculty with appointments in the University of Maryland Institute for Advanced Computer Studies (UMIACS). On those teams, students are applying algorithm design to challenges in computational biology, artificial intelligence and parallel computing.

For Erin Molloy, assistant professor of computer science with an appointment in UMIACS, Chen is helping develop algorithms for phylogeny reconstruction—the process of rebuilding evolutionary relationships using genomic DNA data.

These computational pipelines often amplify small errors, reducing the accuracy of evolutionary trees. Chen's work explores ways to make those algorithms more resilient to uncertain data.

“Computational biology is entirely new to me, but applying algorithm design to genetic data has been an incredible challenge,” Chen said. “Right now, my main focus is doing good research and finding ways to make these evolutionary trees more accurate.”

For Molloy, bridging the gap between artificial intelligence and computational biology offers a fresh perspective on stubborn scientific problems.

"Owen has brought to our group his enthusiasm for artificial intelligence,” Molloy said. “He has been working closely with Ph.D. student a junior majoring in computer science at Harvey Mudd College, is exploring a different challenge under Jordan Boyd-Graber, professor of computer science with an appointment in UMIACS.

Their team studies the strengths and limitations of large language models through Quiz Bowl, a fast-paced trivia competition in which players answer increasingly specific clues. By comparing human and AI performance, the researchers hope to better understand where today's AI systems excel—and where they still struggle.

Nguyen said working in Boyd-Graber's lab immersed him in Maryland's collaborative research environment.

Other REU-CAAR students are working with Laxman Dhulipala, assistant professor of computer science with an appointment in UMIACS, to design algorithms that process enormous datasets more efficiently. Their work explores how computers can divide complex problems across many processors while reducing the time spent moving data between memory and storage—an increasingly important challenge as modern datasets continue to grow.

These projects represent just a sample of the work taking place across REU-CAAR. Gasarch has seen that breadth firsthand while mentoring eight students working on Ramsey theory, a branch of mathematics that studies how order emerges within large, complex systems.

That sustained student interest has helped make REU-CAAR one of the nation's longest-running REU programs, operating continuously since 2013.

“The students keep getting better every year,” Gasarch said. “There are more research opportunities now at a lower level that feed into this program. We're seeing more high school and early college students engaging in research before they even get here.”

For Chen and Nguyen, the experience has made graduate school feel more attainable.

“Before coming here, the idea of a Ph.D. felt incredibly abstract and distant,” Nguyen said. “But being trusted to drive a real project in a top-tier lab changes your perspective.”

—Story by Melissa Brachfeld, UMIACS communications group

Remembering G.W. “Pete” Stewart (1940-2026)

Samuel Malede Zewdu — Thu, 09 Jul 2026 15:07:42 +0000

Stewart, a Department of Computer Science faculty member from 1974 to 2008, worked in numerical linear algebra and helped develop LINPACK, a software package used in scientific and engineering computation.

July 09, 2026

G.W. “Pete” Stewart, a former faculty member in the University of Maryland Department of Computer Science whose work focused on numerical linear algebra and mathematical software, passed away on June 9, 2026.

Stewart joined UMD in 1974 and remained a full-time professor until his retirement in 2008. In addition to the Department of Computer Science, he held appointments in the Applied Mathematics Program, the . After retiring, he continued to be involved in research with the department and was named Distinguished University Professor Emeritus.

Stewart received his bachelor’s and doctoral degrees from the University of Tennessee. Before coming to UMD, he taught at the University of Texas at Austin and Carnegie Mellon University.

His research centered on matrix computations, including matrix decompositions, least squares problems, eigenvalue problems, singular value decomposition and perturbation theory. These areas are part of the numerical methods used in scientific computing, engineering analysis and applied mathematics.

Stewart was one of the developers of

CS Ph.D. student Manan Suri developed Frames2LoRA, a zero-token approach that helps AI analyze videos more efficiently for summarization, question answering and real-time decision-making.

July 08, 2026

Manan Suri, a UMD computer science Ph.D. student, is developing Frames2LoRA to make video AI more efficient.

As artificial intelligence systems take on larger tasks, the amount of information they process has become a practical concern. More tokens can mean higher costs for users, slower responses and greater demand for computing infrastructure, including data centers. Video adds another layer to that issue because a few seconds can contain hundreds or thousands of visual details that a model must process before answering a question or making a decision.

, Suri’s

“My research tries to improve agents to perform better by giving them better context,” Suri said. “The end goal is to improve the context so that the agent can perform better on the tasks that it ends up doing, which is beneficial to everyone.”

Challenge of video

Video is difficult for AI systems because it contains information across time and space. A video is made up of frames, and each frame includes visual information across height and width. As videos get longer or sharper, token counts can rise quickly because the model must represent more frames and more visual detail in each frame.

That growth affects both performance and cost. More tokens require more computation, which can make large models more expensive to run and slower to respond, especially when the same video must be reviewed for multiple questions.

Suri said the issue is not only the amount of data but also how that data affects reasoning.

“When you give AI too much context, it’s not able to reason through it,” Suri said.

The problem is often described as “context bloat,” when information fills a model’s context window, making it harder for the system to identify what matters. Suri compared it to information overload.

The same challenge appears when AI systems repeatedly review video data.

“This essentially has two consequences,” Suri said. “For a long video, the model has too many visual tokens in its context window, and for a lot of tasks, parsing a huge number of tokens introduces latency that we do not need.”

A different path

Frames2LoRA addresses the problem by mapping video information into the model’s weights through an adapter. Instead of requiring the model to repeatedly view the video through visual tokens, the method creates a reusable representation of the video for later questions.

In the project, a perceiver hypernetwork reads intermediate representations generated as a frozen vision-language model encodes a video. It then predicts a LoRA adapter directly from the video in one forward pass, without the iterative updates used in standard LoRA fine-tuning.

The method was trained using SmolVLM2 models with 500 million and 2.2 billion parameters on video summarization and captioning tasks. According to Suri, Frames2LoRA enabled the same frozen model to answer questions using only the adapter, with no visual tokens in context at query time.

Removing visual tokens from the query-time context can reduce memory use, lower latency and make repeated questions about the same video more efficient. It also leaves more room in the context window for text instructions, prior dialogue or other information needed for the task.

Across the evaluation, the method reduced answer-time visual-token load by up to 1,500 times and improved time to first token by six to 80 times while preserving outputs tied to the video. The project also found that adapters generated from separate video segments could compose in rank space, suggesting a possible path for processing longer videos in chunks.

“What Frames2LoRA did for us is that it gave us a very strong proof of concept that you can map context in a single forward pass,” Suri said.

Toward real-world agents

Efficiency becomes especially important as AI systems move toward tasks that depend on continuous visual input. Robots in homes, factories or other settings may need to process a constant visual stream while responding quickly to changing conditions.

“Videos are very relevant when you look at real-world embodied tasks, such as a robot in your home or a robot in a factory,” Suri said. “Its input is a constant visual stream.”

Suri is now exploring whether the same idea can extend beyond perception. Frames2LoRA showed that video context can be internalized in one pass, but future work may examine whether that internalized context can help AI systems act on what they see.

For Suri, that next step is the central question.

“It’s important to know that we can use this to understand videos,” Suri said. “The next question is whether we can make that understanding actionable once the video has been internalized.”

—Story by Samuel Malede Zewdu, CS Communications

Tiny Insects, Big Ideas for AI

Samuel Malede Zewdu — Tue, 07 Jul 2026 16:56:24 +0000

Yiannis Aloimonos studies how insects navigate and communicate to inform the development of autonomous drones.

July 07, 2026

What can bees teach us about artificial intelligence? Quite a bit, according to University of Maryland researchers developing the next generation of autonomous robots.

Yiannis Aloimonos, a professor of computer science with an appointment in the University of Maryland Institute for Advanced Computer Studies (UMIACS), recently shared how insects are shaping the future of miniature drones in an interview with CGTN America.

By studying how bees and other insects navigate, communicate and move through complex environments, researchers are designing drones that can fly more efficiently and adapt to challenging real-world conditions. The work draws on millions of years of evolution to solve modern engineering problems.

Potential applications range from agriculture to emergency response. Insect-inspired drones could one day help pollinate crops, monitor greenhouses and inspect infrastructure. They could also assist first responders by navigating collapsed buildings or other hazardous environments during search-and-rescue operations.

The team is also developing AI systems that stabilize cameras on insect-inspired flying robots, allowing them to collect clearer visual information while in flight. Improving how these drones perceive and respond to their surroundings could make them more effective in environments where traditional aircraft struggle.

For Aloimonos and his collaborators, insects provide an efficient model for building smarter, more capable autonomous systems that could expand the role of AI-powered robotics in everyday life.

—To watch a video of the work click

Ph.D. student Annika Shankwitz, advised by Naomi Feldman, uses machine learning and computational modeling to study how people perceive unfamiliar speech sounds.

July 06, 2026

When learning a new language, one of the biggest challenges is training the ear to recognize unfamiliar sounds. A subtle shift in a vowel or an unfamiliar consonant blend can completely change a word's meaning, yet those distinctions can be difficult for non-native speakers to perceive.

University of Maryland doctoral student Naomi Feldman, a professor of linguistics and an affiliate associate professor of computer science, Shankwitz uses machine learning and computational modeling to develop models that predict how people perceive nonnative speech sounds, helping researchers better understand the relationship between artificial intelligence and human cognition.

“Broadly, I'm interested in how listeners perceive non-native speech sounds,” Shankwitz said. "I want to use machine learning and computational modeling to try and mimic human perceptual patterns that we know about, and by doing that, figure out what computational methods can capture and what they can't."

Earlier this year, Shankwitz received a National Science Foundation (NSF) Graduate Research Fellowship, one of the nation's most prestigious awards for graduate students. The fellowship provides five years of funding for her research exploring computational optimal transport—a mathematical framework for finding the most efficient way to compare complex data—as a model for predicting second-language speech perception.

Shankwitz was one of only six linguistics students nationwide to receive the fellowship this year.

While her NSF-funded research focuses on optimal transport, another branch of her work uses small neural networks to explore how machine learning models speech perception. She will present those findings this month at the Society for Computation in Linguistics (SCiL) workshop, held alongside the Annual Meeting of the Association for Computational Linguistics (ACL 2026) in San Diego.

The

Alan Zaoxing Liu, Christopher Kauffman, Mario Alvarenga, Segev Elisha-Cliff Elazar Mittelman and Amrit Magesh were recognized for teaching and instructional support during the 2025–26 academic year.

July 06, 2026

The University of Maryland’s Department of Computer Science announced its 2025–26 teaching award recipients, recognizing faculty members, teaching assistants and a student-led course facilitator for their work in undergraduate education.

This year’s awardees are Assistant Professor Alan Zaoxing Liu, who received Best Professor; Senior Lecturer Christopher Kauffman, who received Best Lecturer; Segev Elisha-Cliff Elazar Mittelman, who received Best Graduate TA; and Matthias Zwicker, who holds the Elizabeth Iribe Chair for Innovation and the Phillip H. and Catherine C. Horvitz Professorship. “This year’s awardees brought that commitment to their classrooms, discussion sections and student-led courses. We are proud of their contributions to teaching excellence in our department.”

About the Award Recipients

Alan Zaoxing Liu, an assistant professor of computer science, was recognized as Best Professor. Liu joined the department in fall 2023 after serving on the faculty at Boston University

His research spans computer systems, networks and cybersecurity, with a focus on scalable and trustworthy approximate computing systems. At UMD, Liu teaches CMSC414 and CMSC498B, courses that introduce students to computer and network security, cloud computing and applied projects using commercial cloud tools.

“I am honored to receive this award, especially because teaching has been one of the most inspiring parts of my work at UMD,” Liu said. “It is rewarding to see students grow in confidence, take ownership of difficult ideas and surprise themselves with what they are capable of building and understanding. I am grateful to my students, teaching assistants, colleagues and cloud industry partners, whose support and resources have helped turn our educational vision into something meaningful.”

Christopher Kauffman, a senior lecturer in the Department of Computer Science, was named Best Lecturer. He joined UMD in fall 2023 after teaching in the Department of Computer Science and Engineering at the University of Minnesota, Twin Cities.

At UMD, Kauffman teaches CMSC216 and CMSC330, courses in computer systems and programming languages that introduce students to topics such as machine architecture, operating systems and the organization of programming languages.

“I am honored that my teaching has had a positive impact,” Kauffman said. “I have always believed in communities of learning, where our academic mission is strengthened by the people who come together to support it. I am grateful to work alongside outstanding teaching assistants, staff, colleagues and students who help make the classroom successful, and I hope to continue being part of that collective for years to come.”

Mario Alvarenga, a computer science major with a concentration in machine learning, received the Best Undergraduate TA award for supporting undergraduate instruction in CMSC131 and helping students work through course material.

“I am grateful for this award and appreciate the recognition,” Alvarenga said. “It reflects my passion for my students and my commitment to their education. I will continue to push myself and my students toward greater academic success.”

Segev Elisha-Cliff Elazar Mittelman, a computer science Ph.D. student, was recognized as Best Graduate TA. He is advised by Associate Professor of Computer Science Leonidas Lampropoulos, and his research interests include programming languages, formal verification and software testing.

“I am honored to receive this award and grateful to the students who reached out, worked through challenges and spent countless hours trying to understand complex material,” Mittelman said. “Teaching is one of my greatest passions. Few things are more fulfilling than helping a student reach the moment when an idea clicks. Mentorship from great teachers helped me discover my interest in formal methods and testing research, and I hope to help inspire that same curiosity and confidence in the next generation of students.”

Amrit Magesh, a computer science major with a minor in the

—Story by Samuel Malede Zewdu, CS Communications

David Jacobs Named Director of UMIACS

Samuel Malede Zewdu — Wed, 01 Jul 2026 20:19:33 +0000

Jacobs, a computer science professor, will lead the multidisciplinary research institute beginning Sept. 6, 2026.

July 01, 2026

Computer Science Professor David Jacobs was named director of the Amitabh Varshney, dean of UMD’s College of Computer, Mathematical, and Natural Sciences (CMNS). “His record of collaboration across disciplines, combined with his leadership experience, will further strengthen UMIACS’ position as a cornerstone of the university’s research enterprise.”

Jacobs will lead the multidisciplinary research institute where scientists harness the power of computation to tackle some of the most complex challenges facing science and society. The institute brings together more than 130 faculty members from 16 departments who support more than 200 graduate students. UMIACS secures annual sponsored research funding exceeding $30 million in areas such as trustworthy artificial intelligence, machine translation, autonomous robotics, microbiome science, quantum information science, computer vision and cybersecurity.

“My own research has benefited tremendously from the great interdisciplinary environment provided by UMIACS, and I am looking forward to contributing to that environment as director,” Jacobs said.

A leading scientist in the fields of computer vision and machine learning with a focus on visual object recognition and handling lighting variations, Jacobs has worked on a wide range of interdisciplinary projects, collaborating with researchers in electrical engineering, applied mathematics, botany, psychology, chemistry, public policy and more. Jacobs also explores image clustering, structure-from-motion, and the intersections of vision with user interfaces and graphics.

He has published over 160 papers in refereed journals, conferences and workshops and holds five U.S. patents. He has supervised more than 50 high school, undergraduate and graduate students and postdoctoral fellows.

Jacobs has held many leadership roles on campus, including director of the (2007-10). He also led the development of the

The 2026 promotions recognize faculty members whose work spans artificial intelligence, cybersecurity, computational biology and other areas of computing.

June 29, 2026

The University of Maryland Department of Computer Science announced 11 faculty promotions, recognizing faculty members whose work spans artificial intelligence, systems, computational biology, human-computer interaction, security, programming languages, computer vision and computer science education.

The promotions, effective July 1, 2026, include four faculty members promoted to professor, five to associate professor, one to principal lecturer and one to senior lecturer. The faculty members teach and conduct research in areas that support the department’s academic programs and research activity.

“Our promoted faculty have made significant contributions towards building one of the nation’s leading computer science programs through influential research, dedicated teaching and mentorship,” said Department Chair Matthias Zwicker, who holds the Elizabeth Iribe Chair for Innovation and the Phillip H. and Catherine C. Horvitz Professorship. “They are advancing fields such as artificial intelligence, systems and theory while helping educate more than 3,500 undergraduate majors and 600 graduate students. I couldn't be more proud of the achievements of our colleagues.”

Promoted to Professor

Abhinav Bhatele, who has appointments in the (AIM), has been promoted to professor. His research focuses on systems and networks, including parallel computing, large-scale data analytics, programming models, runtimes, network design and simulation, machine learning for parallel systems, and performance modeling and optimization. Bhatele received his Ph.D. in computer science from the University of Illinois Urbana-Champaign in 2010.

Marine Carpuat, who has appointments in UMIACS and AIM, has been promoted to professor. Her research focuses on natural language processing and machine translation, with an emphasis on AI systems that support cross-language communication. Her work examines multilingual AI and the technical and human challenges of using translation tools in real-world settings. Carpuat received her Ph.D. in computer science from the Hong Kong University of Science and Technology in 2008.

Michelle Mazurek, director of the and the Rob Patro has been promoted to professor. He has appointments in UMIACS, ECE and the Leonidas Lampropoulos, who has an appointment in UMIACS, has been promoted to associate professor. His research uses programming-language abstractions to make software and specifications easier to write, debug and reason about. His work focuses on random testing, formal verification and methods for improving software reliability and correctness. Lampropoulos received his Ph.D. in computer science from the University of Pennsylvania in 2018.

Zhicheng Liu, who has an appointment in AIM, has been promoted to associate professor. His research focuses on human-computer interaction, data visualization, graphics, virtual and augmented reality, and machine learning. He directs the Christopher Metzler has been promoted to associate professor. He has appointments in UMIACS, AIM and ECE. His research develops systems and algorithms for computational imaging and sensing, machine learning and wireless communications. His work combines machine learning with imaging and signal processing to study how machines capture and interpret visual and sensory data. Metzler received his Ph.D. in electrical and computer engineering from Rice University in 2019.

Huaishu Peng, who has appointments in UMIACS and AIM, has been promoted to associate professor. His research explores human-computer interaction, design and fabrication, virtual and augmented reality, and assistive technology. His work focuses on digital and physical interactions that make computing more tangible and accessible. Peng received his Ph.D. from Cornell University in 2019.

Rachel Rudinger has been promoted to associate professor. She has appointments in UMIACS and AIM. Her research focuses on natural language processing, including language understanding, computational semantics, commonsense reasoning, and fairness and social bias in AI systems. Her work examines how language technologies understand meaning and operate in real-world settings. Rudinger received her Ph.D. in computer science from Johns Hopkins University in 2019.

Promoted to Principal Lecturer

Mohammad Nayeem Teli, who has an appointment in AIM, has been promoted to principal lecturer. His research focuses on computer vision and machine perception, machine learning and data science. His work centers on systems that use those methods to address computing challenges. Teli received his Ph.D. in computer science from Colorado State University in 2013.

Promoted to Senior Lecturer

Stevens Miller has been promoted to senior lecturer. His interests include computer graphics, especially procedural modeling for the run-time creation of three-dimensional environments for computer games. His earlier work examined segmented architecture machines for computing large synthetic images. Miller received his M.S. from Stevens Institute of Technology in 1988.

—Story by Samuel Malede Zewdu, CS Communications

UMD to Launch 2 Bachelor’s Degrees to Explore Technical, Human Dimensions of AI

Samuel Malede Zewdu — Tue, 23 Jun 2026 18:53:33 +0000

New majors aim to ethically advance AI tech, prepare students to address Its societal impact.

June 23, 2026

The University of Maryland will launch two new undergraduate degrees in artificial intelligence (AI), including one of the nation’s first interdisciplinary majors focused on the impact of AI on humans, with courses spanning philosophy, ethics, public policy and more.

And with a second degree focused on computational structures for AI systems, UMD will continue to build on its more than 60-year history developing the foundations of AI technology, plus a

Meet Yasmine Tajeddin, the Programmer of the Opera

Samuel Malede Zewdu — Tue, 23 Jun 2026 17:55:33 +0000

The senior computer science and music double major never performed publicly before coming to UMD. Now, she’s eyeing a career as a professional opera singer.

June 23, 2026

Unlike most music majors at the University of Maryland, Yasmine Tajeddin had never performed in front of a crowd before arriving on campus. She never had vocal lessons, took a music class in high school or sang in a choir. That’s because in Iran, where she grew up, women can’t sing in public.

“If I wanted to major in something like music, I would need to go outside my country,” she said. So, during her senior year of high school, she decided to audition for the UMD William Gasarch—for her successes and achievements. Now, she’s assisting in instruction herself as a TA for CMSC 351: “Algorithms,” and in past semesters, has been a TA for CMSC 250: “Discrete Structures” and a grader for Professor Research Experience for Undergraduates in Combinatorics, Algorithms, and AI for Real Problems in the Department of Computer Science, working on natural language processing.

“This research is language-related, so I’m really excited about it,” she said.

Balancing science with music has been demanding. Tajeddin’s two majors have no overlapping course requirements. Plus, without extensive musical training, she had to catch up with her classmates who spent high school performing in choirs and orchestras. Between private vocal lessons, practice, choir and participating in the is a mile away on the eastern margin.

“That’s why I had to get a bike,” Tajeddin said. “There’s no way I could get between classes without it.”

Music Professor Emerita and at UMD. For Tajeddin, it was a life-changing experience.

“When I walked on stage for the first time, and I saw this whole orchestra and this amazing venue, it hit me,” she said. “I was like, ‘Oh my god. My first experience with a symphony orchestra wasn’t from the audience—it was from the stage.”

Tajeddin still has one more semester before she graduates. She wants to become a professional opera singer, but through her TA experiences, she learned that she loves teaching, too. So, she’s considering going to graduate school to become a college professor in music or computer science—or maybe even both. To Tajeddin, it's all just language—and whether it’s Java or Japanese, she’s committed to mastering as many as she can.

—Story by CMNS

Can AI and People Play Nice?

Samuel Malede Zewdu — Mon, 22 Jun 2026 18:17:51 +0000

Jordan Boyd-Graber uses a fast-paced trivia competition to study how people decide when to trust AI and when to rely on their own judgment.

June 22, 2026

Artificial intelligence has rapidly transformed from a novelty into a powerful tool capable of outperforming even expert humans in some knowledge-based tasks. In “Can AI and People Play Nice?,” Terp explores how University of Maryland computer science professor Jordan Boyd-Graber is using the fast-paced trivia game quizbowl to better understand the evolving relationship between people and AI.

Boyd-Graber, who has an appointment in the University of Maryland Institute for Advanced Computer Studies (UMIACS), has spent more than a decade developing QANTA, an AI system designed to answer quizbowl questions. In a 2019 competition between top human trivia champions and QANTA, people still held a slight edge, particularly on questions requiring creativity, context and lateral thinking. But as AI technology has advanced, systems like QANTA now outperform even highly skilled human competitors on many measures of raw accuracy.

Rather than focusing on AI-versus-human competition, Boyd-Graber’s research now examines how humans and AI can work together effectively. His studies reveal a critical challenge: AI systems are often overly confident in their answers, even when they are wrong, while humans frequently struggle to determine when they should trust AI recommendations and when they should rely on their own judgment.

Through collaborative quizbowl experiments, researchers found that successful human-AI partnerships depend not only on accuracy but also on trust, transparency and self-awareness. People sometimes reject correct AI answers because the system cannot explain its reasoning, while in other cases they defer to incorrect AI suggestions despite having the right answer themselves.

The findings have implications far beyond trivia. As AI becomes increasingly integrated into fields such as health care, finance and decision-making, understanding how people and machines can collaborate responsibly may be just as important as making AI smarter.

Read the full story:

Computer science Ph.D. student Simge Tekin led the study, which analyzed more than 750,000 container images and found gaps that can leave cloud systems exposed.

June 22, 2026

Simge Tekin (left), Yonghwi Kwon and Tudor Dumitraș recently presented their research on cloud security vulnerabilities at the IEEE Symposium on Security and Privacy, one of the field's premier cybersecurity conferences.

A software patch may be available, but that doesn't mean it's protecting the systems that need it.

A new study from the University of Maryland and Google Research reveals that critical security updates often become trapped in the software supply chains that power modern cloud computing, leaving known vulnerabilities exposed for weeks—or even permanently.

The research team analyzed more than 750,000 software container images—the standardized digital packages used to deploy modern applications—over six years to understand how security fixes move through the cloud ecosystem. Their findings challenge a common assumption that patches automatically flow from software vendors to the applications built on top of their products.

“Many people assume that once a vulnerability is patched, the problem is solved,” said Yonghwi Kwon, an assistant professor of electrical and computer engineering at UMD. “Our research shows that getting that patch to the systems that need it is often the harder problem."

The study was led by Simge Tekin, a third-year doctoral student in computer science at UMD. Her research supervisors, Kwon and Tudor Dumitraș, served as co-authors on the paper. Both Dumitraș, an associate professor of electrical and computer engineering, and Kwon are core members of the University of Maryland Institute for Advanced Computer Studies (UMIACS).

Their paper, was recently presented at the

Researchers found that abandoned upstream layers accounted for 11% of permanently unresolved vulnerabilities. In more complex software dependency chains, that figure rose to 23%.

"When developers realize they aren't receiving critical updates, they often struggle to determine who is responsible for fixing the problem," Kwon said. "The ecosystem lacks visibility into where the update pipeline has broken down."

To address the issue, the team introduced the concept of lineage, which creates a cryptographically verifiable record of a container's ancestry and maintenance history. The approach allows developers to identify software foundations that remain connected to active update channels and avoid dependencies that have effectively been abandoned.

“Our approach essentially builds the missing delivery pipeline,” Kwon said. “Before this, the ecosystem was like a hospital that had the right medicine in the pharmacy, but lacked any functional way to get it up to the patients’ rooms.”

The findings have growing implications for critical infrastructure and other sectors that increasingly depend on cloud technologies. From power grids and water systems to government services and businesses, organizations rely on software that must be updated quickly as new vulnerabilities are discovered. At the same time, advances in artificial intelligence are helping attackers identify and exploit newly disclosed weaknesses faster than ever, increasing the risks associated with delayed patching.

"Understanding the state of security in our critical infrastructure is the first step in defending it," Kwon said.

The research team also included Ph.D. ’21, now a researcher at Google Research. After years of large-scale data collection and analysis, the team's selection for presentation at S&P highlights both the significance of the findings and the potential of the lineage approach to strengthen software security practices across the cloud computing industry.

—Story by Melissa Brachfeld, UMIACS communications group

CS Ph.D. Student Eadom Dessalene and Postdoc Levi Burner Earn Intelligence Community Fellowships

Samuel Malede Zewdu — Sat, 20 Jun 2026 14:43:04 +0000

June 20, 2026

Years after beginning their research careers in the same lab, Eadom Dessalene (left) and Levi Burner (right) are launching fellowship-supported projects that explore how intelligent systems perceive and interpret the physical world. They are joined by Yiannis Aloimonos (center left) and Christopher Metzler (center right).

Two University of Maryland researchers will use prestigious Intelligence Community (IC) Postdoctoral Research Fellowships to tackle challenges ranging from atmospheric turbulence to human-machine understanding.

Eadom Dessalene, who will complete his Ph.D. in computer science this summer, have been selected for the highly competitive program, which is funded primarily by the Office of the Director of National Intelligence and supports basic research in areas of strategic interest to the U.S. intelligence community. The fellowships begin in September 2026 and support two years of research.

Both conducted doctoral research in the Yiannis Aloimonos, a professor of computer science with an appointment in the University of Maryland Institute for Advanced Computer Studies (UMIACS). The group's work emphasizes machine perception, embodied intelligence and learning from real-world interaction.

Though their research paths have diverged, both scholars are using artificial intelligence (AI) to solve a common challenge: helping machines better understand the physical world. Through their fellowships, Burner and Dessalene will pursue projects that combine advanced sensing technologies with AI, enabling systems to interpret information that is difficult or impossible to observe directly.

“Levi and Eadom represent the kind of researchers we strive to develop at Maryland—scientists who are willing to tackle fundamental questions while also pursuing applications with real-world impact,” Aloimonos said. “Their work approaches machine intelligence from different directions, but both are pushing the boundaries of how systems learn from and understand the world around them. Having two of our researchers selected for these fellowships is a tremendous recognition of their accomplishments and potential.”

Seeing Through Atmospheric Turbulence

Burner currently serves as a postdoctoral fellow in both the Maryland Robotics Center and the Christopher Metzler, an assistant professor of computer science with an appointment in UMIACS.

Levi Burner (left) and Christopher Metzler discuss research aimed at predicting atmospheric turbulence before it distorts images or disrupts optical communications. Their fellowship-supported work combines advanced imaging technologies and AI to forecast atmospheric fluctuations in real time.

Burner and Metzler will collaborate with a researcher from the U.S. intelligence community to develop new approaches for predicting atmospheric turbulence using advanced imaging systems and AI.

Atmospheric turbulence—the rapidly changing fluctuations in air that distort the path of light—poses a major challenge for long-range imaging and laser-based communication systems. Rather than correcting those distortions after they occur through expensive post-processing, Burner's team aims to predict them on microsecond timescales, allowing systems to compensate for turbulence before it degrades images or communications.

To accomplish this, the researchers will combine two emerging imaging technologies. Single-photon avalanche diode (SPAD) cameras can detect individual photons and capture imagery at microsecond speeds, making them uniquely suited to observing fast-changing atmospheric conditions. The team will also use light-field imaging, which captures multiple views of a scene simultaneously along different atmospheric paths. By comparing these simultaneous views, researchers can separate the chaotic fluctuations of the air from actual motion in the underlying environment when imaging moving targets.

AI will play a critical role in analyzing these massive streams of data and identifying patterns that can be used to forecast atmospheric fluctuations.

“Atmospheric turbulence is one of the most significant barriers to obtaining clear imagery and reliable optical communication over long distances,” Metzler said. “By combining new sensing technologies with machine learning, we're exploring whether these distortions can be predicted before they occur, opening the door to entirely new capabilities in imaging and communication.”

The work could have applications ranging from high-speed, long-range imaging systems to next-generation optical communications in space.

“Much of my research has focused on helping machines build meaningful representations of the world through observation and interaction,” Burner said. “This project applies those ideas to an environment that is constantly changing. We're asking whether AI can learn the underlying structure of turbulence well enough to predict it and improve how imaging systems perceive the world.”

Helping AI Understand Human Actions

Dessalene's fellowship research will continue his collaboration with Aloimonos. The two will pair their expertise with a researcher from the intelligence community to focus on a challenge that remains difficult for even the most advanced AI systems: understanding how humans physically interact with objects.

Yiannis Aloimonos (right) discusses concepts in embodied intelligence and machine perception with Eadom Dessalene. Building on research that began in the Perception and Robotics Group, Dessalene's fellowship project uses first-person video and muscle activity data to help AI better understand how people interact with the physical world.

While modern video models can often recognize broad activities, they frequently struggle to understand the lower-level hand movements, action primitives and physical forces exchanged during object manipulation. Dessalene's project seeks to bridge that gap by building AI systems that can learn structured hierarchies of action from both visual and physiological data, developing higher-level understanding from these low-level sub-actions.

His research will combine egocentric, or first-person, video with electromyography (EMG) data collected from wrist-worn sensors. EMG measures the electrical activity produced by muscles, providing critical information about hand movements, contact and force that may be invisible in video alone.

By pairing video with EMG signals, Dessalene is training deep learning models to essentially “see the forces” exchanged between hands and objects. The approach could help AI systems develop a more physically grounded understanding of human actions, moving beyond visual observation alone.

“Eadom’s research tackles one of the most fundamental abilities of the human cognitive system—how we learn by watching others,” Aloimonos said. “What sets his work apart is the way it combines visual information with signals from the body itself. By capturing muscle activation alongside video, he is creating a technology with applications spanning healthcare, education, defense and robotics.”

The fellowship builds on Dessalene's doctoral research in robot learning, embodied AI, egocentric vision and multimodal perception. His previous work explored how robots can learn from human videos and how physical interaction can improve visual representations of actions and objects.

In one project, Dessalene helped develop a large-scale dataset that paired first-person video with force measurements collected during natural object manipulation tasks, creating new opportunities for researchers to study the relationship between visual observations and physical interactions.

“Humans learn an enormous amount simply by observing others and interacting with the world,” Dessalene said. “My goal is to help AI systems move beyond recognizing what someone appears to be doing and toward understanding how they are physically engaging with their environment. That kind of understanding could have important applications in robotics, assistive technologies and augmented reality.”

A Shared Foundation

While Burner and Dessalene are pursuing different scientific questions, both trace their research roots to the Perception and Robotics Group's emphasis on machine perception and embodied intelligence.

Their fellowship projects—one focused on understanding atmospheric turbulence and the other on understanding human action—reflect a shared goal of helping machines better understand the physical world.

Their selection for the IC Postdoctoral Research Fellowship highlights the impact of that research environment and underscores the growing role of UMD researchers in advancing AI, robotics, sensing and machine learning in areas of national importance.

—Story by Melissa Brachfeld, communications group

Mohammad Hajiaghayi Earns ACM SIGecom Test of Time Honors

Samuel Malede Zewdu — Thu, 04 Jun 2026 14:31:40 +0000

June 04, 2026

Mohammad Hajiaghayi, the Jack and Rita G. Minker Professor of Computer Science at the University of Maryland, will be recognized with two Test of Time Awards from the Association for Computing Machinery’s Special Interest Group on Economics and Computation (

Mohammad Hajiaghayi will be recognized with two ACM SIGecom Test of Time Awards, one for each of two papers that helped shape the fields of algorithmic economics and online mechanism design.

The awards will be presented at the upcoming 27th ACM Conference on Economics and Computation (EC’26), held this year in Rome from July 6–10.

“These papers were ahead of their time in identifying fundamental challenges in online markets and proposing techniques that remain influential today,” said ” published in the proceedings of the 2004 ACM Conference on Electronic Commerce (EC’04). Hajiaghayi co-authored the paper while he was a Ph.D. student at the Massachusetts Institute of Technology alongside

The paper addressed a central challenge in online markets: how to design auctions when buyers arrive over time and the seller has only a limited supply of goods. It pioneered the use of the “secretary problem” in online auctions and mechanism design, where buyers arrive sequentially in random order and allocation decisions must be made immediately without knowledge of future arrivals.

The authors studied settings in which bidder valuations were drawn independently from unknown distributions and developed adaptive mechanisms that learn from early arrivals before deciding when to allocate resources. More broadly, the work introduced secretary-style methods into online mechanism design, creating a framework for balancing exploration and commitment under uncertainty.

The resulting mechanisms achieved provable guarantees for revenue and social welfare, helping influence later work in online mechanism design, matching and real-time market algorithms.

The research also combined ideas from online algorithms, auction theory and economics to create mechanisms that remained “strategyproof”—meaning participants could not gain an advantage by misrepresenting information—while still performing well compared with an ideal offline auction. The paper additionally established mathematical limits on how well online mechanisms could perform relative to an optimal offline auction, results that later became influential in research on online decision-making and mechanism design.

At the time, online auctions and digital marketplaces were becoming increasingly important in areas such as advertising, ticket sales and e-commerce. The paper helped establish theoretical foundations for understanding how markets function when decisions must be made in real time without knowing future demand.

The second award recognizes “

The research addressed a fundamental challenge in digital markets: how to design online auction mechanisms when sellers have statistical information about buyers but do not know the eventual size of the market.

The work pioneered the use of “prophet inequalities” in mechanism design, introducing one of the first frameworks showing how prior distributional information can be used to make near-optimal online decisions under uncertainty.

Prophet inequalities compare an online algorithm—which must make irrevocable decisions sequentially—to an idealized “prophet” with complete knowledge of future outcomes. The paper showed how historical distributions of bidder valuations could be used to derive adaptive pricing and allocation rules that closely approximate the performance of this offline optimum.

The work established a framework for using historical statistical information to design effective online pricing and allocation mechanisms, influencing later research in online mechanism design, posted-price systems, revenue optimization and learning-augmented algorithms.

The paper also proved strong impossibility results, showing that without reliable information about market size, no online mechanism can consistently achieve a constant-factor approximation to optimal revenue or social welfare.

Together, the two papers helped establish foundational frameworks for online mechanism design by introducing secretary and prophet methods into online auctions and economic decision-making. Their ideas influenced later work in online matching, pricing, allocation and market design.

ACM SIGecom, which presents the award, promotes research at the intersection of economics and computation, with a focus on using computational reasoning and economic theory to better understand markets, incentives and social interactions in digital environments.

The Test of Time Award is among the most selective honors in economics and computation, recognizing papers that continue to influence research and applications years after publication. Past recipients include researchers whose work defined major areas of theoretical computer science and online markets, including landmark papers on Nash equilibria, digital goods auctions and online advertising systems.

Hajiaghayi joined the University of Maryland faculty in 2010, where his interview talk focused on the two papers now receiving Test of Time recognition. Since then, he has built an internationally recognized research program spanning algorithms, artificial intelligence, game theory, machine learning and optimization. His work has addressed topics ranging from network design and computational complexity to market mechanisms and AI for social good.

In addition to his appointment in the Department of Computer Science, Hajiaghayi holds a joint appointment with the University of Maryland Institute for Advanced Computer Studies (UMIACS) and is a core faculty member of the

UMD Team Develops Precise “Undo Button” for AI Memory

Samuel Malede Zewdu — Fri, 29 May 2026 15:38:16 +0000

Led by Soheil Feizi, researchers developed a method that can remove specific information from AI models without diminishing their broader abilities.

May 29, 2026

Imagine trying to remove a single drop of red dye from a gallon of purple paint without ruining the color entirely. For developers of large language models (LLMs), that has long been the challenge of “unlearning”—the process of removing specific information from an AI system after it has already been trained.

Once sensitive personal information, copyrighted text or harmful misinformation becomes embedded in an AI model, it spreads across billions of internal connections. Until now, the most reliable solution was often the most extreme: discard the model and retrain it from scratch, a process that can cost millions of dollars and consume enormous amounts of energy.

Researchers at the University of Maryland and the Max Planck Institute for Software Systems have developed a new alternative—a method that functions like a surgical rollback tool for AI memory.

Led by Soheil Feizi, an associate professor of computer science with an appointment in the University of Maryland Institute for Advanced Computer Studies (UMIACS), the team created a framework that can selectively erase unwanted information from an AI model while preserving its overall intelligence and reasoning skills.

Their study, introduces a technique called Model State Arithmetic (MSA). The approach uses saved training snapshots—known as checkpoints—to identify exactly how problematic data altered a model during training and then reverse those changes with remarkable precision.

“We realized that the model’s own development history contains all the information we need to fix it,” said Feizi, who is also affiliated with the “By using these training snapshots as a guide, we can identify exactly how specific data points altered the AI’s behavior and simply reverse those changes. It’s a much more elegant solution than trying to force a finished model to forget through trial and error.”

Most existing unlearning methods attempt to make a completed AI model forget information after the fact, often degrading the system’s broader capabilities in the process. Researchers sometimes describe the side effects as “brain damage,” where models lose general knowledge, reasoning ability or even coherent language generation.

MSA takes a different approach by revisiting earlier stages of the model’s training process. During training, developers routinely save checkpoints as safeguards against crashes or hardware failures. Feizi and his collaborators discovered those same snapshots could be repurposed as a historical record of how information became encoded inside the model.

The system identifies a “clean” checkpoint from before the model encountered the data targeted for removal. Researchers then briefly retrain that earlier version on the unwanted material to measure how the model changed. The resulting mathematical signature—what the team calls a “forget vector”—captures the influence of the problematic data. Subtracting that vector from the final model effectively rolls back the unwanted information while leaving the rest of the system intact.

The researchers tested MSA using industry benchmarks designed to evaluate machine unlearning, including the Task of Fictitious Unlearning (TOFU) and the Machine Unlearning Six-Way Evaluation (MUSE). In experiments involving fictional biographies and copyrighted books, the method consistently removed targeted information while preserving the model’s reasoning and conversational performance.

Remarkably, the team found that the technique remained effective even when using checkpoints saved very early in training, suggesting that even infrequent backups may provide enough information to support precise unlearning.

As governments and technology companies face mounting pressure to comply with privacy laws, copyright protections and emerging “right to be forgotten” standards, the work could offer a practical path forward for AI developers seeking to update or correct models without rebuilding them entirely.

“AI models shouldn’t be permanent black boxes that can never be corrected once they’re trained,” said held last month in Rio de Janeiro. In addition to Feizi and Rezaei, the research team included of the Max Planck Institute for Software Systems. Together, their work advances a growing effort at UMD to develop AI technologies that are not only more powerful, but also more transparent, adaptable and accountable.

—Story by Melissa Brachfeld, UMIACS communications group

Alan Liu Receives NSF CAREER Award to Improve Monitoring of AI and Cloud Systems

Samuel Malede Zewdu — Fri, 29 May 2026 14:13:00 +0000

The five-year project will develop new telemetry methods to help operators monitor large-scale AI and cloud systems more efficiently.

May 29, 2026

Alan Liu, an assistant professor of computer science at the University of Maryland, has received a (CAREER) Program award to advance research on telemetry and observability for large-scale AI and cloud infrastructure.

Liu, who also has appointments in the (MC2) and the

—Story by Samuel Malede Zewdu, CS Communications

Ashok Agrawala to Retire After 55 Years

Samuel Malede Zewdu — Wed, 27 May 2026 21:26:55 +0000

Agrawala’s career spans the department’s early years, decades of work in systems, networking and mobile computing, and the program’s growth into a top-ranked computer science program.

May 27, 2026

When Ashok Agrawala arrived at the University of Maryland in 1971, computer science at the university was still taking institutional shape. The Department of Computer Science had not yet been formally established, the personal computer era had not yet begun and many of the systems that now support daily communication remained research questions.

More than five decades later, Agrawala is preparing to retire after 55 years at UMD, closing a faculty career that began before the department’s founding in 1973 and continued through major changes in computing research, education and infrastructure.

Agrawala, a professor of computer science who previously held appointments in the , has worked in systems, networking, real-time computing, wireless systems, synchronization and health informatics. His research has also extended into machine learning, data science and the Internet of Things.

Reflecting on his time at UMD, Agrawala said his long career has given him a view of the department’s development to its current role as a top-ranked program.

“When I joined UMD, computer science was still a young discipline working to define its foundations and broader role,” Agrawala said. “Since then, the department has grown from a small group of faculty and students into a major center for research, education and innovation. What impresses me most is not just its size or reputation, but its ability to keep reinventing itself while maintaining a commitment to rigorous thinking, strong teaching and meaningful research.”

Department Chair Matthias Zwicker said Agrawala’s career reflects both the department’s history and the long-term role faculty members play in shaping its research and educational mission.

“Ashok’s career at UMD is unique in its scope, spanning more than five decades and reflecting the department’s journey from its founding to the program we see today,” said Zwicker, who holds the Elizabeth Iribe Chair for Innovation and the Phillip H. and Catherine C. Horvitz Professorship. “He has been a steady presence of change in computer science, contributing to the department not only through his research, but also through his tireless mentorship, service and commitment to students.”

Agrawala’s academic career began in India, where he earned bachelor’s and master’s degrees in electrical engineering from the Indian Institute of Science in Bangalore. He later earned a master’s degree and doctorate in applied mathematics from Harvard University, completing his Ph.D. in 1970 before joining UMD the following year.

At UMD, his career became closely tied to the department’s early development. The department was formed from the Computer Science Center in 1973, and its Bachelor of Science degree program began that same year.

As the department grew, Agrawala’s work followed computing into increasingly networked and mobile environments. One of his widely cited contributions is the Glenn Ricart (Ph.D. ’80, computer science), for mutual exclusion in distributed systems. The algorithm addresses how multiple computers in a distributed system can coordinate access to a shared resource without interfering with one another.

That work reflected an area of computing that became increasingly important as networks expanded beyond individual machines. Over time, Agrawala’s research moved from coordination problems in distributed systems to applications involving wireless devices, real-time information and public safety.

“Information has been the central thread in my work, including how it is represented, communicated, interpreted and used to coordinate action,” Agrawala said. “My interests have moved across systems, networks, distributed computing, mobile computing and public safety applications, but the underlying question has remained the same. How can information systems help people and organizations function more effectively? I have always been interested in how technology interacts with real human needs, decisions and consequences, and that broader view has shaped much of my research.”

He later directed the M-Urgency, a wireless campus safety application developed by Agrawala and student programmers. The app allowed users in distress to send audio, video and location information to emergency responders. In 2012, he received two awards for his contributions to public safety through the project, including a VITA Wireless Samaritan Award from The Wireless Foundation and CTIA-The Wireless Association.

The project reflected Agrawala’s interest in connecting research with applications that could address immediate needs. It also gave students a way to work on technology with a clear public safety purpose.

“Working with students has been one of the most important and rewarding parts of my career,” Agrawala said. “Projects such as M-Urgency were shaped by students who challenged ideas, built prototypes and helped keep the work grounded in real-world needs. Public safety is an area where the connection between computing and society becomes very direct, and it reminds us that computer science is not only about abstraction and algorithms, but also about responsibility, usability, reliability and impact.”

His work in wireless systems also included the From Bits to Being: Computation, Consciousness, and the Next Frontier of Computer Science,” examining how computer science has grown into a force shaping human civilization and why the field must also consider the people for whom computational systems are built. For pictures from the event, click

He discusses his path into computing, his work in computational genomics and advice for students interested in computational biology.

May 27, 2026

University of Maryland Associate Professor of Computer Science Rob Patro works in computational genomics, where his research addresses the growing need to process and interpret large amounts of biological sequencing data. He leads the

CS Students Receive Grant Family and Ben Shneiderman Awards

Samuel Malede Zewdu — Wed, 27 May 2026 14:02:48 +0000

The awards recognize undergraduate double majors in computer science and mathematics and doctoral research in human-computer interaction.

May 27, 2026

The University of Maryland’s Department of Computer Science recognized four undergraduate students and one doctoral student with awards honoring work across computer science, mathematics and human-computer interaction.

Recent computer science and mathematics majors (B.S. '26, computer science; mathematics), (B.S. '26, computer science; mathematics) received the Grant Family Outstanding Achievement Undergraduate Student Award in Computer Science and Mathematics. Computer science Ph.D. student Zeyu Yan (Ph.D. ’26, computer science) received the Ben Shneiderman Outstanding Thesis Award in Human-Computer Interaction.

The Grant Family Award, established by Adjunct Professor of Computer Science John Grant, recognizes graduating seniors who pursue double majors in computer science and mathematics. Grant has been a faculty member in both the mathematics and computer science departments for more than 50 years.

The Ben Shneiderman Outstanding Thesis Award in Human-Computer Interaction, named after Professor Emeritus Ben Shneiderman, recognizes research contributions by a master’s or doctoral student whose thesis advances the theory or practice of HCI.

About the Award Recipients

Nishkal Hundia

Hundia began his undergraduate studies as a computer science major, but his interest in artificial intelligence and language models led him to study mathematics alongside it. He said the decision reflected his interest in understanding the mathematical ideas that support many areas of computer science.

“So much of what makes computer science powerful is really a wrapper around core mathematical concepts, and I wanted to understand those ideas at their roots rather than just work with the abstractions on top,” Hundia said.

That perspective shaped how Hundia thinks about emerging areas of research, particularly as AI becomes more visible across scientific fields. He said the technical progress behind those systems is often discussed in broad terms, even though much of the work depends on mathematical reasoning.

For Hundia, studying both disciplines also gave him a way to connect theory with the systems being built around it.

“I’m incredibly honored to receive this award and deeply grateful to everyone who supported me throughout my undergraduate journey,” Hundia said. “Beyond the personal recognition, it feels like a validation of something I believe strongly: that computer science is best understood when studied alongside the mathematics that underlies it, and I’m glad that’s something worth celebrating.”

Anirudh Satheesh

Satheesh was first drawn to computer science through an interest in building practical applications that could solve real-world problems. His coursework later led him to pursue a deeper mathematical understanding of the systems and algorithms behind those applications.

Satheesh said mathematics and computer science will continue to shape research in connected ways. He said mathematics can help define the limits of what is possible, while computer science can help determine what can be implemented at scale. He pointed to artificial intelligence, quantum computing and other computational sciences as areas where both disciplines will remain important.

Receiving the Grant Family Award, Satheesh said he recognized the value of pursuing both fields together.

“It is especially meaningful because it recognizes the value of pursuing both computer science and mathematics together,” Satheesh said. “I am grateful to the faculty, mentors and peers who supported me throughout my time at UMD. The award motivates me to continue pursuing research at the intersection of these disciplines.”

Tahmid Zaman

Zaman’s path into computer science began with programming and the possibility of using software to build tools with relatively few barriers to entry. Mathematics offered a different but related appeal, giving him a way to understand patterns, systems and natural phenomena with precision.

“I had always been interested in programming: the idea that anyone could leverage computers to build something impactful was something that I could be passionate about when I entered university,” Zaman said. “At the same time, I always appreciated how math precisely describes the natural phenomenon and dynamics of our world, and how even the most complex mathematical abstractions have some real-world application.”

The connection between the two fields became a defining part of his academic path. For Zaman, computer science offered tools for building and testing ideas, while mathematics provided the language for understanding the structures behind them.

“The intersection between these fields made it an obvious choice to study both,” he said.

Zaman said the relationship between computer science and mathematics is becoming increasingly visible in scientific research, particularly in areas that rely on large datasets, predictive models and advanced computing systems. He pointed to protein structure prediction, quantum computing, post-quantum cryptography and AI-assisted formal verification as examples of areas shaped by both disciplines.

As he prepares to move beyond his undergraduate studies, Zaman said the award came at a meaningful point in his academic path.

“I am truly grateful to receive the award, and especially thankful to UMD’s professors, clubs and other resources that supported me on my journey,” Zaman said. “It inspires me to apply the knowledge and experiences I’ve gained as I begin my career.”

Zeyu Yan

Yan, advised by Assistant Professor of Computer Science Huaishu Peng, received the Ben Shneiderman Outstanding Thesis Award in Human-Computer Interaction for research examining how computation can be integrated into the physical world. His work centers on HCI, with an emphasis on how computational methods can shape materials, artifacts and creative processes.

“My research sits at the intersection of computation and the physical world,” Yan said. “I have always enjoyed building things, and over time I became fascinated by how computation can be embedded into the processes, materials and artifacts we create.”

In Yan’s view, computation can change the pace and scope of research by helping researchers model systems, test ideas and refine designs before moving into more resource-intensive stages of experimentation. For work that crosses disciplines, those tools can make it possible to explore problems that do not fit neatly within one field.

Receiving the Shneiderman award, Yan said, was meaningful not only as recognition of his thesis work but also as support for research that crosses academic boundaries.

“I feel incredibly honored and grateful,” Yan said. “Beyond the recognition itself, I deeply appreciate Ben and the broader community for investing their time and resources in supporting researchers who work across disciplinary boundaries.”

Yan said interdisciplinary research can sometimes fall between traditional fields, making recognition from the HCI community especially significant. To him, the award also reflects the role computer science can play in connecting technical research with other domains.

—Story by Samuel Malede Zewdu, CS Communications

UMD CS Celebrates More Than 1,100 Graduates

Samuel Malede Zewdu — Tue, 26 May 2026 13:46:45 +0000

CS students who graduated with bachelor’s, master’s and doctoral degrees in August 2025, December 2025 and May 2026 were honored at last month’s commencement ceremonies.

May 26, 2026

For many students, commencement is measured in brief moments: a name called from the stage, a handshake, a photo with family or a final walk with classmates. Behind those moments are years of coursework, research, teaching, mentoring and career preparation.

The University of Maryland Department of Computer Science recognized those milestones during the 2026 , who was advised by Assistant Professor of Computer Science Huaishu Peng, the department’s size and range of research areas played a central role in shaping his graduate experience.

“I think it is actually a real honor to be in a program, first of all, that is big, and not only by population but also by disciplines,” Yan said. “Nowadays, everyone, including me, is doing interdisciplinary research, and you cannot expect one person to do everything that solves a huge sociotechnical issue.”

Yan said his time at UMD shifted how he thinks about research, moving from learning technical skills to identifying problems and using those skills to address them.

“I came in the program carrying all the fundamental knowledge and skills,” Yan said. “More importantly, what I achieved is now I can look at the world slightly differently through training, so that I know I can identify things that need to be solved and know what I can do to resolve those questions.”

That perspective will continue to guide Yan after graduation. He plans to pursue a postdoctoral fellowship as he works toward a long-term goal of becoming a faculty member. He said his doctoral training prepared him to identify societal issues, develop research questions and work toward outcomes that can be published or applied beyond the university.

For undergraduate No. 7 among public institutions. Its 2026 undergraduate program ranked No. 9 among public universities, with notable placements in artificial intelligence, cybersecurity and computer systems. A recent University Career Center survey also showed that Department of Computer Science graduates achieved a 93% job placement.

As graduates move from the classroom, lab and campus community into new professional paths, Department Chair Matthias Zwicker said they are prepared to adapt to a changing field.

“Our graduates leave UMD with the technical foundation, curiosity and resilience needed to make meaningful contributions in a rapidly changing field,” said Zwicker, the Elizabeth Iribe Chair for Innovation and the Phillip H. and Catherine C. Horvitz Professor. “As they move into the workforce, graduate study and new opportunities, we wish them continued success and look forward to seeing the impact they will make as lifelong CS Terps.”

For more pictures from the 2026 Commencement Ceremonies, click

Her research examines how irrelevant information can influence reasoning in large language models.

May 18, 2026

University of Maryland Department of Computer Science Ph.D. student Hillary Owusu received an , where she will present research examining irrational reasoning behaviors in large language models (LLMs). ACL 2026 is an international conference focused on natural language processing and computational linguistics.

Owusu, who is advised by Affiliate Associate Professor of Computer Science Naomi Feldman, studies how AI systems can be influenced by irrelevant numerical information, a phenomenon related to the anchoring effect in cognitive science.

Her research focuses on improving the reliability of systems that power tools such as ChatGPT. Her paper examines how anchoring appears in language models, how it relates to model uncertainty and how post-training methods can affect model behavior.

“As AI systems become more involved in how people learn, work and make decisions, their reliability has real societal consequences,” Owusu said. “My paper shows that even small, irrelevant details in a prompt can influence a model’s judgment in ways that users may not notice from the final answer alone.”

Research on LLMs has increasingly examined issues related to reasoning reliability, bias and interpretability as AI systems become more widely used in education, health care, business and other sectors. Researchers have also explored how prompting methods and training techniques shape the outputs and decision-making patterns of these systems.

Owusu said identifying and measuring anchoring effects could contribute to the development of AI systems that are more transparent and dependable in settings where accuracy and trust are important.

“Making these effects visible and measurable is an important step toward building AI tools that people can use with greater confidence,” Owusu said.

According to Owusu, the ACM-W scholarship provides a valuable opportunity to present her research to peers with similar interests in the field.

“I am deeply grateful to receive this award,” Owusu said. “This recognition is especially meaningful because it helps make it possible for me to attend ACL 2026 and share my research with the broader AI community.”

Owusu said the recognition also encourages her to continue research connecting computational cognitive science and natural language processing.

The work is part of Owusu’s broader focus on model reliability.

“My future work will focus on moving from diagnosis to intervention, not only identifying when anchoring occurs in large language models, but also understanding how it is represented internally and how its negative effects can be reduced,” Owusu said. “This work brings together computational cognitive science and AI interpretability to support the development of AI systems that are more reliable in real-world use.”

—Story by Samuel Malede Zewdu, CS Communications

UMD Researchers Take on AI’s Hardest Problem: Teaching Machines to Reason

Samuel Malede Zewdu — Mon, 18 May 2026 15:42:52 +0000

Mohammad Hajiaghayi and CS Ph.D. students are developing AI systems that can generate, test and refine mathematical ideas.
May 18, 2026
At the University of Maryland, researchers aren’t just building smarter AI—they’re trying to teach machines how to reason.

Led by computer science professor Mohammad Hajiaghayi, a UMD team is working toward an ambitious goal: AI systems that can generate, test and refine mathematical ideas on their own. Backed by a
UMD computer science Ph.D. students (from left) Iman Gholami, Arshia Soltani, and Danny Mittal are collaborating on a project to build AI systems that don’t just solve problems—but learn to reason through them.

At the center of the effort are three computer science Ph.D. students—and University of Maryland Institute for Advanced Computer Studies (UMIACS), where Hajiaghayi has an appointment, allows the team to test ideas quickly—an advantage he says has accelerated several research directions.

The implications extend beyond speed.

If AI systems can reliably understand problems, reason through them and verify solutions, they could reshape the structure of research. Instead of producing one-off answers, these systems could revisit questions, refine ideas over time and contribute continuously—less like tools, more like collaborators.

The project is high-risk. Core questions about reasoning, formalization and reliability remain unresolved, and there’s no guarantee AI can meet the rigor required for true mathematical discovery.

But that uncertainty is part of the point, the researchers say.

For this team, the goal isn’t just better AI—it’s a different role for it in science. If they succeed, research may no longer be a solitary pursuit, but a collaboration between human intuition and machine reasoning.

And in that shift, the line between tool and thinker may begin to blur.

—Story by Stratis Aloimonos, UMIACS communications group

UMD’s Putnam Team Finishes in Top 15 for Ninth Straight Year

Samuel Malede Zewdu — Mon, 18 May 2026 13:41:52 +0000

CS majors Sujay Konda and Ishan Raghavendra each ranked in the top 200 out of more than 4,300 competitors in one of the most prestigious mathematics contests for undergraduates in North America.
May 18, 2026
Every winter, thousands of college students across the United States and Canada sit down for six hours to take the 12-question Putnam Competition exam, widely considered one of the most difficult undergraduate math contests in the world.

Most people who tackle the test walk away with a score of zero. Getting even a single point is considered an accomplishment.

But in December 2025, three University of Maryland students—computer science majors and mathematics major Daniel Yuan—represented UMD as a team and individually finished in the top 200 out of over 4,300 Putnam competitors.

Overall, UMD placed 14th among 487 participating institutions—marking the ninth consecutive year that the university has finished in the top 20 nationally. Although not part of UMD’s official team, computer science major Hanson Bai and mathematics majors Boheng Shao and Kelin Zhu were each recognized among the top 500 as well.

For Doron Levy, chair of UMD’s Department of Mathematics, this year’s results mean something beyond just a ranking.

“Schools like MIT, Princeton and Stanford recruit math majors specifically for their Putnam teams,” Levy said. “This is an amazing achievement, in particular given that there are very few public universities with comparable ranking. Our continued success says a lot about the quality of our students and the support they receive here.”

First-timers at the top

What makes this year’s results particularly striking is that two of the students who represented UMD in the competition had never tackled the Putnam before. Unlike Yuan, a returning contestant with previous Putnam success, Konda and Raghavendra were first-time competitors. Yet both freshmen finished among the top scorers in the country.

Raghavendra, who plans to add a mathematics major, arrived at UMD already well-versed in competition math after years of proof-based contests, including high school-level American Mathematics Competitions, American Invitational Mathematics Examinations and HMMT tournaments. For him, Putnam was less a leap into the unknown than a continuation of something he already loved.

“I heard about the contest in high school and decided to participate because I love proof-based math,” Raghavendra said. “I credit most of my preparation to my previous math competition background, but Putnam was still a very different experience.”

With a background rooted more in competitive programming, Konda heard about the competition through friends and decided to give it a shot. He went in curious, worked through old Putnam exams to prepare and came out with something he had not quite expected—a new way of seeing problems.

“Coming from a mostly computer science background, I think taking the Putnam helped expand the different perspectives and ways to approach problems from a different field,” he said. “It was interesting to see the different types of solutions that exist in math, stuff you don’t see in computer science.”

The formula for success

Students like Konda and Raghavendra are exactly the kind of contestants Roohollah Ebrahimian, a principal lecturer in UMD’s Department of Mathematics and the university’s Putnam coordinator, wants to reach. Ebrahimian, who has coached UMD’s Putnam team for more than a decade, noted that UMD’s nine-year winning streak was no accident.

“The formula for our success really comes down to two things,” he said. “Recruitment of our best students and extensive preparation.”

Ebrahimian maintains a running list of close to 100 students—drawn partly from top finishers at a high school math competition he runs for students in D.C. and Maryland (Yuan and Zhu both participated). Ebrahimian personally reaches out to each student at the start of every fall semester.

“UMD’s one-credit, 90-minute Putnam Express course has also helped prepare decades of students for the contests,” noted Ebrahimian, who teaches the class and established its accompanying guide. “The class predates my own time here and ensured many high scorers.”

Ebrahimian also found less conventional ways to stay connected to students, such as offering office hours in UMD’s dining halls to chat informally with students directly over a meal.

A few years ago, he also launched a YouTube channel to reach UMD math majors and prospective students and support their academic aspirations. Within weeks, students were finding his channel and emailing him about his math videos, and what began as an outreach experiment grew into something more. Ebrahimian began uploading videos specifically about Putnam and how to tackle math competition exams. The videos work through multiple failed attempts before landing on an answer, with him narrating what went wrong at each turn and explaining the “behind-the-scenes” reality of competitive test-taking.

“I wanted to be a resource for anyone trying to understand not just what the solutions to competition problems were but also how to arrive at them,” Ebrahimian said. “There are many resources that give you problems along with solutions, but not many that actually walk you through the process.”

Thanks to his efforts both online and in real life, UMD’s reputation for mathematical rigor and success in competitions has spread widely. In fall 2025, students from other institutions—including local community colleges in Maryland—began sitting in on Ebrahimian’s Putnam Express classes, hoping to find a formula for success.

“They’ve seen our success and have seen how Maryland has been consistently doing well,” Ebrahimian said. “Students want to be a part of that and learn.”

What’s next

Encouraged by their results this year, Konda and Raghavendra look forward to the next Putnam competition in December. Both freshmen are eager to test themselves against the nation’s best undergrad mathematicians again. And they hope others join them—the more, the merrier.

“Putnam’s for anyone passionate about math,” Raghavendra said. “There’s nothing to lose because even just getting a non-zero score is a big deal, and you can always try again the next year.”

“Considering how much fun I had preparing for Putnam and the different problem-solving skills I’ve developed from it, I think I’ll probably continue to participate for the rest of the time I’m here at UMD,” Konda added. “I just hope that more students will get involved in one way or another.”

—Story by CMNS

UMD CS News

When AI Goes on Defense

6 Faculty Members Join UMD’s Department of Computer Science

NSF Renews Maryland-Led Quantum Simulation Institute’s Funding

UMD Course Connects Computer Science, Math and Sports Analytics

The Code They Couldn't Talk About

CS Researchers Receive AIM Seed Grant to Study the Roots of AI Bias

Noriko Maitland Receives UMD Provost’s Academic Advisor of the Year Award

CMMRS Marks 10 Years of Connecting Students With CS Research

TRAILS Awards More Than $515K for Trustworthy AI Research

Training the Next Public Health Workforce

Safer AI for Autonomous Drones

Helping Local Governments Govern AI

CS Faculty Join Federal Initiative to Speed AI-Powered Discovery

Learning When to Pivot

Making Science Easier to See

Making Complex Math as Easy as Asking a Question

Three CS Terps Received CMNS Alumni Network Endowed Undergraduate Awards

Remembering G.W. “Pete” Stewart (1940-2026)

Tiny Insects, Big Ideas for AI

David Jacobs Named Director of UMIACS

UMD to Launch 2 Bachelor’s Degrees to Explore Technical, Human Dimensions of AI

Meet Yasmine Tajeddin, the Programmer of the Opera

Can AI and People Play Nice?

CS Ph.D. Student Eadom Dessalene and Postdoc Levi Burner Earn Intelligence Community Fellowships

Seeing Through Atmospheric Turbulence

Helping AI Understand Human Actions

A Shared Foundation

Mohammad Hajiaghayi Earns ACM SIGecom Test of Time Honors

UMD Team Develops Precise “Undo Button” for AI Memory

Alan Liu Receives NSF CAREER Award to Improve Monitoring of AI and Cloud Systems

Ashok Agrawala to Retire After 55 Years

CS Students Receive Grant Family and Ben Shneiderman Awards

About the Award Recipients

UMD CS Celebrates More Than 1,100 Graduates

UMD Researchers Take on AI’s Hardest Problem: Teaching Machines to Reason

UMD’s Putnam Team Finishes in Top 15 for Ninth Straight Year