Back in 2011, a team of University of Washington and University of California San Diego researchers published a paper detailing how they could remotely hack into a pair of 2009 Chevy Impalas. By targeting a range of attack vectors including CD players, Bluetooth and cellular radio, the researchers were able to control multiple vehicle functions, from the windshield wipers to the brakes.
Since its publication, the team’s research has helped lead to new standards for motor vehicle security and put the brakes on automobile cyberattacks. For their lasting contributions, their paper titled “Comprehensive Experimental Analyses of Automotive Attack Surfaces” received the Test of Time Award at the 34th USENIX Security Symposium in Seattle earlier this month.
Franziska Roesner
“I was only a first-year graduate student when we started this project, and I had just switched my focus to security. It was such a privilege to be able to help out on such an important and impactful project, and to learn from all of the other members of the team about how to do this kind of research,” said co-author Franziska Roesner (Ph.D., ‘14), Brett Helsel Professor and co-director of the Security and Privacy Research Lab in the Allen School.
Modern automobiles are made up of independent computers called electronic control units (ECUs), typically connected through the Controller Area Network (CAN), that oversee different motor functions. In a previous paper, the team found that if an attacker physically connected to the car’s internal network could override critical safety systems. Building off of that work, the researchers analyzed the modern automobile’s external attack surface and found that an adversary could hack into a car from miles away.
The team identified three categories of components that were vulnerable to cyberattacks. An attacker could use an indirect physical channel such as tools that connect to the OBD-II port, which can access all CAN buses in the car, or through the media player. For example, the researchers compromised the car’s radio and then used a doctored CD to upload custom firmware. If an attacker is able to place a wireless transmitter in proximity to the car’s receiver, they can gain access to the ECU via Bluetooth or even remote keyless entry, the team found. Attackers do not have to be nearby to wreak havoc. Using long-range communication channels such as cellular, it is possible to exploit vulnerabilities in how the car’s telematics unit uses the aqLink code to remotely control the vehicle.
“More than 10 years ago, we saw that devices in our world were becoming incredibly computerized, and we wanted to understand what the risks might be if they continued to evolve without thought toward security and privacy,” said senior author Tadayoshi Kohno, who was then a professor at the Allen School, now faculty at Georgetown University, in a UW News release.
The impact of the team’s work can still be felt today. As a result of the research, car manufacturers including GM have hired entire security teams. The work has influenced the development of guidelines for original equipment manufacturers (OEMs) and also led to the creation of the Electronic Systems Safety Research Division at the National Highway Traffic Safety Administration. As cars grow increasingly more connected and autonomous, the insights from the UW and UCSD collaboration will continue to inform the automotive industry against emerging threats.
“Beyond the practical impact of the work, that experience has also made for great stories to tell in the computer security courses I teach now — for example, the time that we accidentally set the car’s horn to a permanent ‘on’ state while experimenting outside the Allen Center,” Roesner said.
Joining Roesner and Kohno at UW at the time of the original paper were Karl Koscher (Ph.D. ‘14), now a postdoc at UCSD, and Alexei Czeskis (Ph.D., ‘13), currently at LinkedIn. The original University of California San Diego group was made up of UCSD faculty members Stefan Savage (Ph.D., ‘02) and Hovav Shacham; Stephen Checkoway (B.S., ‘05), now faculty at Oberlin College; Damon McCoy, faculty at New York University; Danny Anderson, who runs a software consulting company; and late researcher Brian Kantor.
The Allen School at the University of Washington is working with Ai2 and other partners on a new initiative to advance open AI for science and the science of AI, with support from the U.S. National Science Foundation and NVIDIA.
The University of Washington’s Paul G. Allen School of Computer Science & Engineering has teamed up with the Allen Institute for AI (Ai2) on a new project aimed at developing the first fully open set of artificial intelligence tools to accelerate scientific discovery and enhance the United States’ leadership in AI innovation. Today the U.S. National Science Foundation (NSF) and NVIDIA announced a combined investment of $152 million in this effort, including $75 million awarded through the NSF’s Mid-Scale Research Infrastructure program.
Ai2 will lead the Open Multimodal AI Infrastructure to Accelerate Science (OMAI) project. The principal investigator is Ai2 Senior Director of NLP Research Noah A. Smith, who is also Amazon Professor of Machine Learning at the Allen School. Smith’s faculty colleague Hanna Hajishirzi, Torode Family Professor at the Allen School, is co-principal investigator on behalf of UW and also Ai2’s senior director of AI.
“OMAI is a terrific opportunity to leverage the longstanding partnership between Ai2 and the Allen School, which has yielded some of the most exciting developments in building truly open AI models and trained some of the most promising young scientists working in AI today,” said Hajishirzi. “This is a pivotal moment for us to form the foundation for scientific discovery and innovation across a variety of domains — and also, importantly, advance the science of AI itself.”
Noah A. Smith (left) and Hanna Hajishirzi aim to leverage the partnership between Ai2 and the Allen School to benefit science and society.
The cost of building and maintaining today’s AI models is too prohibitive for all but the most well-resourced companies, leaving researchers in academic and not-for-profit labs without ready access to these powerful tools and stifling scientific progress. The goal of the OMAI project is to build out this foundational infrastructure through the creation and evaluation of models trained on open-access scientific literature and informed by the needs of scientists across a range of disciplines. By openly releasing the model weights, training data, code and documentation, the team will provide researchers using its tools with an unprecedented level of transparency, reproducibility and accountability, instilling confidence in both the underlying models and their results.
The concept for OMAI was incubated in an ecosystem of open research and collaboration that the Allen School and Ai2 have built since the latter’s founding in 2014. That ecosystem has enabled dozens of UW students to collaborate with Ai2 on research projects, produced leading-edge open AI artifacts like the Open Language Model (OLMo) and Tulu, and developed tools like OLMoTrace to give anyone full visibility into models’ training data — all of which have helped fuel Seattle’s emergence as a hub of AI innovation.
Smith looks forward to leveraging that longstanding synergy to push technologies that will have a transformational impact on the American scientific enterprise — and even transform the conversation around AI itself.
“There’s been a reaction that seems to be widespread that AI is a thing that is happening to us, as if we are passively subject to this technology and don’t have agency,” Smith said. “But we do have agency. We get to define what the priorities should be for AI and to build tools that scientists will actually be able to use and trust. With OMAI, the UW will be a leader in this new paradigm and push AI in a more responsible direction that will benefit society in a multitude of ways.”
In addition to the UW, academic partners in the OMAI project include the University of Hawai’i at Hilo, the University of New Hampshire and the University of New Mexico.
OMAI represents a landmark NSF investment in the technology infrastructure needed to power AI-driven science — a development that Brian Stone, performing the duties of the agency’s director, described as a “game changer.”
“These investments are not just about enabling innovation; they are about securing U.S. global leadership in science and technology and tackling challenges once thought impossible,” Stone said.
Understanding how different parts of the brain communicate is like trying to follow conversations at a crowded party — many voices overlap, some speakers are far away and others might be hidden entirely. Neuroscientists face a similar challenge: even when they can record signals from multiple brain regions, it is difficult to figure out who is “talking” to whom and what is being said.
Matt Golub
In a recent paper published at the 2025 International Conference on Machine Learning (ICML), a team of researchers led by Allen School professor Matt Golub developed a new machine learning technique to cut through that noise and identify communication between brain regions. The technique, called Multi-Region Latent Factor Analysis via Dynamical Systems (MR-LFADS), uses multi-region neural activity data to decode how different parts of the brain talk to each other — even when some parts can’t be directly observed.
“The many regions within your brain are constantly talking to each other. This communication underlies everything our brains do for us, like sensing our environment, governing our thoughts, and moving our bodies,” said Golub, who directs the Systems Neuroscience & AI Lab (SNAIL) at the University of Washington. “In experiments, we can monitor neural activity within many different brain regions, but these data don’t directly reveal what each region is actually saying — or which other regions are listening. That’s the core challenge we sought to address in this work.”
Unlike existing approaches, MR-LFADS is able to automatically account for unobserved brain regions. For example, neuroscientists can use electrodes to simultaneously monitor the activity of large populations of individual neurons across multiple brain regions. However, this activity may be influenced by neurons and brain regions that are not being recorded, explained Belle Liu, UW Department of Neuroscience Ph.D. student and the study’s lead author.
“Imagine trying to understand a conversation when you’re not able to hear one of the key speakers. You’re only hearing part of the story,” Liu said.
To overcome this, the team devised a custom deep learning architecture to detect when a recorded region reflects an unobserved influence and to infer what the unobserved region was likely saying.
“We wanted to make sure the model can’t just pipe in any unobserved signal that you might need to explain the data,” said co-author and Allen School postdoc Jacob Sacks (Ph.D., ‘23). “Instead, we figured out how to encourage the model to infer input from unobserved sources only when it’s very much needed, because that information can’t be found anywhere else in the recorded neural activity.”
The team tested MR-LFADS using both simulated brain networks and real brain data. First, they designed simulated multi-region brain activity that reflected complicated scenarios for studying brain communication, such as giving each region unique signals from both observed and unobserved sources. For the model, the challenge is to recover those signals and to disentangle the source of those signals and whether they come from observed regions — and if so, which ones — or unobserved regions. The researchers found that their model was able to infer this communication more accurately than existing approaches. When applied to real neural recordings, MR-LFADS could even predict how disrupting one brain region would impact another — effects that it had never seen before.
By helping neuroscientists better map brain activity, this model can help provide insights into treatments for various brain disorders and injuries. For example, different parts of the brain communicate in certain ways in healthy individuals, but “something about that communication gets out of whack in a diseased state,” explained Golub. Understanding when and how that communication breaks down might enable the design of therapies that intervene in just the right way and at just the right time.
“Models and techniques like these are desperately needed for basic neuroscience to understand how distributed circuits in the brain work,” Golub said. “Neuroscientists are rapidly improving our ability to monitor activity in the brain, and these experiments provide tremendous opportunities for computer scientists and engineers to model and understand the intricate flow of computation in the brain.”
The RoboFly (left) in comparison to the TinySense sensor (center) next to a pencil for scale.
Flying insect robots (FIRs) have the potential for use in search and rescue operations, environmental monitoring and even space missions due to their small size and low material cost. The challenge, however, is finding the minimum sensor suite and computation resources, or avionics, needed for the robot to maintain flight and control.
A team of researchers in the University of Washington’s Autonomous Insect Robotics (AIR) Lab developed TinySense, the current lightest avionics system with the potential for FIR sensor autonomy. Smaller than the size of a penny and less than half the size of the previous lightest avionics system, TinySense features a global shutter camera, a gyroscope and a pressure sensor to help the FIR estimate the different variables needed to control hover — pitch angle, translational velocity and altitude. The team presented their research titled “TinySense: A Lighter Weight and More Power-efficient Avionics System for Flying Insect-scale Robots” at 2025 IEEE International Conference on Robotics and Automation (ICRA) and received the Best Student Paper Award.
“Despite huge progress towards flying insect robots like the UW’s RoboFly and Harvard’s RoboBee, none have yet been able to fly using only sensors carried onboard,” said co-lead author and Allen School undergraduate student Joshua Tran. “The TinySense is light and efficient enough to finally make this feat a possibility, and opens the door to many other tiny flying applications like the TinyQuad and Coincopter, gram-scale propeller drones also from our lab.”
The TinySense sensor is smaller than the size of a penny.
TinySense builds on and improves previous FIR sensor suites from the AIR Lab to create an avionics system that is even better tailored in mass and energy consumption for an insect-scale robot. To help reduce the system’s mass and power needs, the team first replaced the power-hungry laser rangefinder with a lighter and more efficient Bosch BMP390 pressure sensor. They then replaced the bulky optic flow sensor with a novel global shutter camera and a custom-written optic flow algorithm running on a 10 milligram microcontroller — small enough to fly onboard an FIR. TinySense weighs approximately 75 milligrams and uses about 15 milliwatts of power to fly.
“The team made important contributions in a number of areas that hadn’t previously been addressed because nobody has been thinking deeply about how to make flight controllers really efficient and lightweight. They built a new ultra-light flex circuit, their own camera optics and then performed extensive validation on the full system they created,” said senior author Sawyer Fuller, UW Department of Mechanical Engineering professor and Allen School adjunct faculty member.
The team demonstrated the TinySense sensor suite onboard the Crazyflie, the smallest commercially available sensor-autonomous flying robot, and found that TinySense had a comparable performance to the industry-standard sensors on the Crazyflie. In future work, the team aims to integrate TinySense into Robofly so that it will be able to, for the first time, hover without needing external sensors.
(From left to right) Joshua Tran, Claire Li and Zhitao Yu earned a Best Student Paper Award for TinySense at ICRA.
“It was exciting to hear the interest in the TinySense project and its future integration with the Robofly at the ICRA conference,” said co-author and Allen School undergraduate student Claire Li.
For co-lead author and mechanical engineering Ph.D. student Zhitao Yu, working on TinySense also gave him the opportunity to help mentor the next generation of researchers.
“Mentoring Josh and Claire was a rewarding experience on this project,” said Yu. “It was great to see them grow into confident researchers and contribute meaningfully to such a challenging and impactful system.”
The cardinality estimation problem, or the challenge of accurately predicting the size of the output to a query without actually evaluating the query, is one of the oldest and most important problems in databases and data management. Cardinality estimation helps guide decisions on every aspect of query execution, from how much memory should be allocated for storing the query result to the number of servers needed to successfully process an expensive query. However, cardinality estimation is notoriously difficult; current methods can often have large errors, leading to poor decisions downstream.
Dan Suciu
A team of researchers led by Allen School professor Dan Suciu of the UW Database Group introduced a new pessimistic cardinality estimator called LpBound which provides a guaranteed upper bound on the query output size. This method offers a strong, theoretical guarantee that for any database that meets the given statistics, the query output size will always be below the bound set by LpBound. They presented their research titled “LpBound: Pessimistic Cardinality Estimation using Lp-Norms of Degree Sequences” at the 2025 ACM SIGMOD/PODS International Conference on Management of Data last month and received a Best Paper Award for their work.
“Cardinality estimation is difficult, because it needs to rely on a very small amount of information (statistics on the input data), and needs to produce an accurate estimate,” said senior author Suciu, who also holds the Microsoft Endowed Professorship in the Allen School. “The novel solution described in the paper estimates the cardinality of the output by using simple statistics on the input data, and applying Shannon inequalities from information theory. The method outperforms not only traditional cardinality estimators, but also novel estimators based on machine learning.”
The LpBound cardinality estimator provides several advantages over other learned estimators currently available, including FactorJoin, BayesCard and DeepDB. In addition to the guaranteed upper bounds, it has a low estimation time and error as well as space requirements, making it useful for practical applications. LpBound also works for both cyclic and acyclic queries — meaning it can estimate the cardinality in traditional SQL workloads, which are often acyclic, and in graph pattern matching or SparQL queries, which are more likely to be cyclic. When integrated into the query optimization framework PostgreSQL, the researchers found that LpBound’s estimates led to query plans as good as those made from true cardinalities, making it more applicable for real-world database systems.
Magdalena Balazinska, professor and director of the Allen School, has been elected a member of the Washington State Academy of Sciences (WSAS) in recognition of her “contributions in data management for data science, big data systems, cloud computing, and image/video analytics and leadership in data science education.” The WSAS was established in 2015 as a source of independent, evidence-based scientific and technical advice for state policy makers, modeled after the National Academies of Science, Engineering and Medicine. Balazinska, who was directly elected by her WSAS peers, is one of 36 members in the 2025 class.
“We are pleased to recognize the achievements of these world-renowned scientists, engineers, and innovators,” said WSAS President Allison Campbell. “And we are grateful for their willingness to contribute expertise from a wide range of fields and institutions to support the state in making informed choices in a time of growing complexity.”
One of Balazinska’s most influential achievements has been her foundational work on Borealis, a distributed stream processing engine that made large-scale, low-latency data processing more dynamic, flexible and fault tolerant for a variety of applications, from financial services and industrial processing, to network monitoring and wireless sensing. Borealis introduced the ability to quickly and easily modify queries at runtime in response to current conditions, correct query results to account for newly available data, and allocate resources and optimize performance across a variety of networks and devices. Earlier this year, Balazinska and her collaborators earned a Test of Time Award at the Conference on Innovations Data Systems Research (CIDR 2025) for their work on Borealis. They received a Test of Time Award in 2017 from the Association for Computing Machinery Special Interest Group on the Management of Data (ACM SIGMOD) for a related paper expanding the system’s fault tolerant stream processing capabilities.
Balazinska also advanced the then-burgeoning field of “big data,” particularly for scientific applications. She co-led the design and development of Myria, a fast, flexible, open-source cloud-based service that enabled domain experts across various scientific fields to perform big data management and analytics. Myria was designed for efficiency and ease of use; it also functioned as a test-bed for Balazinska and her colleagues to explore new directions in data management research in response to real users’ needs. Her work on Myria and related projects earned Balazinska the inaugural VLDB Women in Database Research Award at the International Conference on Very Large Databases in 2016.
More recently, Balazinska has focused on data management for visually intensive applications such as video and augmented, virtual and mixed reality. For example, she and her collaborators developed VOCAL, or Video Organization and Compositional AnaLytics, to make it easier for users to organize and extract information from any video dataset. In the absence of a pretrained model, the system combines active learning with a clustering technique to reduce the manual effort involved in identifying and labeling features. It also supports compositional queries for analyzing the interaction of multiple objects over time, and it can self-enhance its own capabilities by using large language models (LLMs) to identify and generate missing functionality in response to user workloads.
Balazinska, who has served as director of the Allen School since 2020, holds the Bill & Melinda Gates Chair in Computer Science & Engineering at the University of Washington and is a senior data science fellow in the eScience Institute. She previously served as director of the eScience Institute and associate vice provost for data science at the UW, in addition to co-chairing the National Science Foundation’s Advisory Committee for Computer and Information Science and Engineering (CISE). Last year, Balazinska was appointed to Washington state’s Artificial Intelligence Task Force charged with developing recommendations on potential guidelines or legislation governing the use of AI systems. She currently co-chairs two task-force subcommittees focused on AI in education and workforce development and in health care and accessibility, respectively.
A total of 12 UW faculty members were elected as part of the incoming WSAS class, which also includes Allen School adjunct professor Julie Kientz, chair of the Department of Human-Centered Design & Engineering. Kientz was recognized for her research and leadership in human-computer interaction that “has advanced health and education technology, influenced policy, and shaped the HCI field through impactful scholarship, interdisciplinary collaboration, and inclusive, real-world technology design.” Balazinska, Kientz and their colleagues will be formally inducted at an event marking the Academy’s 20th anniversary in October.
Balazinska is the fourth Allen School faculty member to be elected to the WSAS; professors Anna Karlin and Ed Lazowska and professor emeritus Hank Levy previously joined following their elections to the National Academies of Science and/or Engineering.
Allen School professor Abhishek Gupta, who directs the Washington Embodied Intelligence and Robotics Development (WEIRD) Lab, is interested in developing ways to help robots learn new skills with minimal human help and engineering. Gupta joined the Allen School faculty in 2022, and already he has introduced research that has shaped the future of robotics.
“It is an honor to receive this award, which will support our group’s ongoing research into robot learning methods that are deployable and improvable in high-impact, human-centric environments,” Gupta said.
A robot uses the Cherrybot reinforcement learning system to acquire fine manipulation skills such as picking up a cherry.
Gupta’s research has focused on developing methods that can make it practical for robots to improve safely and reliably through reinforcement learning. However, applying reinforcement learning to real-world robotics presents challenges, from safety, to reward specification, to efficiency; as such, its success has been limited to controlled settings or simulation. To address these challenges, Gupta established that robotic systems learning in the real world need to be able to determine the success measure from its own sensory input, and then reset the environment without human help so it can retry solving a task and learn from a small set of real-world interactions. He subsequently demonstrated what that process looks like via projects focused on dexterous, multi-fingered hands, fine manipulation tasks and teaching robots to grasp different objects through expert demonstrations.
His work was one of the first to propose solutions to the unavailability of automatic resets — one of the most fundamental, yet often overlooked, struggles in implementing robot learning in the real world. He developed a formalism and set of benchmark tasks that help robots navigate a continual, non-episodic world without assuming access to an oracle reset mechanism. His later work addressed the reset-free learning problem as actually being a multi-task learning problem, where a robot performing some tasks then resets others. These systems and algorithms set the stage for the next generation of deployment systems that will not just remain static, but improve autonomously on the job through multi-task reset-free data collection. Gupta has also built a range of reinforcement learning libraries and tooling to make real-world learning accessible to a broader range of developers.
Building off of that research, Gupta has been investigating how leveraging alternative sources of data such as generative models, simulation and videos can help scale up robotic learning. He and his collaborators were one of the first to develop GenAug, short for generative augmentation, which uses a diffusion model to synthetically modify robotic images for improved generalization. This system tackles the issue of the lack of in-domain robotics data through pre-trained generative models.
Using the RialTo system, robots can learn new skills in “digital twin” simulation environments that they can then transfer to the real world. (Dennis Wise/University of Washington)
Gupta has also introduced a new method for robotic learning in simulation by using a real world to simulation and then back to the real world approach. Using small amounts of real-world data, researchers can construct a simulation of the deployment area that the robot can interact with and learn from. Simulations also have helped the design of effective policies that, when deployed in the real world, could help robots perform tasks with many variations and disturbances. For example, robots using this framework can efficiently put away dishes in a dish rack where they have to account for different dish shapes and configurations. Through a set of algorithmic ideas, Gupta and his collaborators were able to directly transfer behaviors from simulations to reality, and then efficiently finetune those behaviors using small amounts of real-world experience. More recently, Gupta and his students have developed techniques for learning unified prediction and control models from raw-video experience, allowing for the use of large internet-scale datasets in robot learning.
In addition to pioneering real-world reinforcement learning, Gupta has developed methods for unsupervised and self-supervised reinforcement learning. By combining the best of both worlds from model-based and model-free reinforcement learning, he introduced a simple and effective self-supervised reinforcement learning technique to make successor representations more practical using deep reinforcement learning methods. These representations help predict how likely a robot is to visit different states in the future. He and his collaborators were also among the first to develop a method that enables robots to learn useful skills without a reward function. This work has prompted a subcommunity of research on unsupervised reinforcement learning and skill discovery for both robotics and machine learning.
Abhishek Gupta (center) poses with members of his WEIRD Lab among their robots. (Dennis Wise/University of Washington)
“Abhishek’s work has been consistently creative, innovative and practical, making a significant impact on the current and future state of robotic reinforcement learning,” Allen School professor Dieter Fox said. “He’s been a wonderful person to collaborate with, and we are very excited to have him in the Allen School. Abhishek’s work is laying the foundation for the next generation of robot learning, and he is poised to become one of the key leaders in our field.”
Prior to earning this year’s RAS Early Academic Career Award, Gupta has received a 2024 Amazon Science Hub Research Award and was named a 2023 Toyota Research Institute Young Faculty Investigator.
Allen School Ph.D. student Cheng-Yu Hsieh is interested in tackling one of the biggest challenges in today’s large-scale machine learning environment — how to make artificial intelligence development more accessible. Large foundation models trained on massive datasets have revolutionized AI, however, these scaling efforts are often out of reach for many except for well-resourced companies, Hsieh explained. His research focuses on making both data and model scaling more efficient and affordable to help democratize AI development.
“I develop effective data curation techniques for training large foundation models, as well as efficient methods to deploy and adapt these models to various downstream applications. My research spans key stages of today’s artificial intelligence development pipeline,” said Hsieh, who works with professor Ranjay Krishna in the UW RAIVN Lab and affiliate professor Alex Ratner in the Data Science Lab.
For his contributions, Hsieh was awarded a 2024 Google Ph.D. Fellowship in machine intelligence. The fellowship recognizes outstanding graduate students from around the world representing the next generation of leaders with the potential to influence the future of technology through their research in computer science and related fields.
“This fellowship will support my research on making large-scale AI systems more efficient, accessible and adaptable. I’m excited to continue exploring how we can make AI technology more sustainable and inclusive,” Hsieh said.
Hsieh designs methods to help mitigate the high costs and other complexities associated with large-scale AI model development. For example, one of the major bottlenecks in today’s machine learning pipeline is manually labeling or curating large datasets, which can be labor intensive. Hsieh and his collaborators introduced Nemo, an end-to-end interactive system that guides users through creating informative datasets using weak supervision techniques in order to lower the barrier for building capable AI models in low-resource settings. Nemo was able to improve overall workflow efficiency by 20% on average compared to other weak supervision approaches, Hsieh found.
Some of his research projects have been put into practice and have already made a real-world impact. As part of a collaboration between UW and Google, Hsieh helped develop the distilling step-by-step method that enables users to train smaller task-specific models using less training data compared to other standard fine-tuning or distillation approaches. With this method, a smaller 770M parameter T5 model trained with only 80% of the data on a benchmark can outperform a much larger 540B PaLM model. The team launched the project on Google Vertex AI, the company’s generative AI development platform, and Google highlighted the research at the 61st Annual Meeting of the Association for Computational Linguistics (ACL 2023). Hsieh’s research into model adaptation was also integrated into the Vertex platform, allowing users to adapt models to new applications without needing explicit training data.
“Cheng-Yu is a self-sufficient, diligent, effective and productive researcher,” said Krishna. “His recent papers propose solutions to a wide range of pertinent problems in natural language processing, efficient machine learning and retrieval augmented generation, and I have no doubt that he will continue to produce impactful research.”
As part of his goal to make data and model scaling more efficient and affordable, in future research, Hsieh is interested in developing new approaches for querying powerful, but oftentimes expensive, generative AI models to help create informative and controllable datasets for model training and alignment.
“This fellowship is both a recognition of the work I’ve done and an incredible encouragement to continue pushing my research direction in AI. I am very thankful to my advisors, mentors and collaborators who have supported me along the way,” Hsieh said. “I am excited to continue pursuing research with real-world impact in this fast-paced era of AI development.“
(left to right) MLSys Program Chairs Celine Lin and Gauri Joshi presented the Best Paper Award to the FlashInfer team members Lequn Chen (Ph.D., ’24), Ruihang Lai, Zihao Ye, Tianqi Chen (Ph.D. ’19) and Luis Ceze.
A team of University of Washington and NVIDIA researchers developed a system that can help make large language models (LLMs) faster and more adaptable. The foundation of LLMs are built on transformers, a neural network architecture driven by attention mechanisms that help artificial intelligence focus on relevant and important information. As these LLMs evolve and find new applications in diverse fields, however, optimized lower-level implementation, or GPU kernels, become necessary to help prevent errors and ensure low-latency inference.
The researchers introduced FlashInfer, a versatile LLM inference kernel library that is open source as well as highly optimized and adaptable to new techniques including key-value, or KV, cache reuse algorithms. They presented their research titled “FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving” at the Eighth Annual Conference on Machine Learning and Systems (MLSys 2025) in May and received a Best Paper Award.
“FlashInfer proves what’s possible when academia, industry and the open-source community innovate together — ideas jump from whiteboard to GPU kernels at lightning speed,” said lead author and Allen School Ph.D. student Zihao Ye, who completed part of the research during his internship at NVIDIA. “That shared, rapid feedback loop lets us iterate, refine and ship breakthrough inference speedups that keep pushing the limits of large language models.”
FlashInfer is able to address major challenges that LLMs face in memory access and heterogenous hardware. The attention engine uses a unified block-sparse format, where data is stored and organized in dense blocks making it easier to navigate, to optimize KV cache storage and composable formats. FlashInfer can also adapt to various attention mechanisms through just-in-time compilation, while the dynamic load-balanced scheduling framework effectively and efficiently handles different workloads. Compared to other state-of-the-art LLM serving solutions, the researchers found that FlashInfer significantly boosted kernel performance across diverse inference scenarios. Already, FlashInfer has been integrated into several leading LLM serving frameworks, including SGLang, vLLM and MLC Engine.
A packed Alaska Airlines Arena celebrates the Allen School’s graduating class of 2025. (Photo by Kerry Dahlen)
On Friday, June 13, an estimated 5,000 friends, family, faculty and staff packed the Alaska Airlines Arena in the University of Washington’s Hec Edmundson Pavilion to celebrate the Allen School’s graduating class of 2025. While the date invited superstition, the evening was full of jubilation as roughly 800 graduates collected their commemorative diplomas, flipped their tassels and made the transition from Allen School students to Allen School alumni.
“It feels like yesterday that we were welcoming you to the Allen School. And tonight we celebrate all you have accomplished here,” said Magdalena Balazinska, professor and director of the Allen School. “We are extremely proud of you. We are proud of all you have accomplished and can’t wait to see what you accomplish next!”
‘True impact doesn’t come from what you accumulate, but what you contribute.’
Trish Millines Dziko: “Changing the world isn’t about being remembered — it’s about doing things worth remembering.” (Photo by Matt Hagen)
That was the message graduation speaker Trish Millines Dziko — co-founder and executive director of the Technology Access Foundation (TAF), computer scientist and proud Husky mom — delivered to the graduates as they contemplate the next stage of their journey. She was welcomed to the stage by professor Ed Lazowska, who exercised one of his last official acts in his 48th and final year as an Allen School faculty member by introducing Millines Dziko, calling her “a friend of mine and a hero of mine.”
“TAF uses STEM as a tool for social change,” Lazowska said during his introductory remarks. “And in its nearly 30 years, TAF has changed the lives of tens of thousands of students in our area.”
Technology as a tool for social change was a recurring theme in Millines Dziko’s speech — but, she noted, not always for the better. With more than 60% of the nation’s wealth concentrated in the top 10% of households, while the bottom 50% hold just 5%, Millines Dziko suggested that the world needs “more people who care enough to fix what’s broken.”
“You can use your critical thinking, problem solving, ideation, creation and leadership skills to build solutions to some of the most pressing problems like homelessness, generational poverty, public education, the environment and health care,” she said.
Whatever path they decide to pursue, the graduates will not be able to rely on their technical skills alone. Saying hard work and good grades were “just the beginning,” Millines Dziko advised the graduates to prioritize building relationships by showing themselves to be capable, reliable, truthful, empathetic and accountable. Developing these qualities would enable them to build social capital that, she noted, they could use along with technology and engineering “as the vehicle to creating a better future for everyone.”
“I hope you pursue purpose over profit, and let your values lead your vision. Please create solutions that lift people up and improve communities,” Millines Dziko urged. “Because in the end, changing the world isn’t about being remembered — it’s about doing things worth remembering.”
Alumni Impact Award: Nicki Dell (Ph.D., ‘15)
Making our computer-mediated world safer and and more equitable: Nicki Dell with Magda Balazinska (left) and Shwetak Patel (Photo by Matt Hagen)
Nicki Dell is a shining example of what Millines Dziko talked about. Each year, the Allen School recognizes one or more alumni who have used their Allen School education to change the world. Since her own graduation a decade ago, the 2025 Alumni Impact honoree has been “doing things worth remembering” in the form of technologies that serve the needs of overlooked communities such as home health care workers and people experiencing intimate partner violence. Dell worked with professors Linda Shapiro and the late Gaetano Borriello on her way to earning the 500th doctoral degree awarded by the Allen School before taking up a faculty position at Cornell Tech.
In his remarks, presenter Shwetak Patel, professor and associate director of development and entrepreneurship, highlighted Dell’s leadership of the Clinic to End Tech Abuse (CETA) among her many contributions — contributions that had already earned her a SIGCHI Societal Impact Award as well as a MacArthur Foundation “genius grant.”
“She deeply partners with affected communities, and then builds systems and interventions that make our computer-mediated world safer and more equitable for everyone,” Patel said.
Recognizing student leadership and service
So many members of the Allen School’s undergraduate student body — which now numbers more than 2,200 — contribute to activities and events that enrich the student experience, it is difficult for the Undergraduate Student Services Team (USST) to choose the recipients of this and the Outstanding Senior awards. But choose, they did; after USST Director Crystal Eney invited all graduating students who were involved in outreach, community building and mentorship to stand and be recognized, the following individuals were singled out for their contributions to the Allen School community and the field of computing.
Undergraduate Service Awards
Inspirational, compassionate and mission-driven: (left to right) Kianna Roces Bolante, Joo Gyeong Kim and Anjali Singh (Photo by Kerry Dahlen)
Honoree Kianna Roces Bolante was described as the “epitome of service” in her role as chair of the student group Computing Community, or COM^2, overseeing school-wide events and activities that build community among the undergraduate majors in the Allen School. In her two years at the helm, she earned a reputation — and universal appreciation — for interacting with the community she serves with empathy, intention and a commitment to inclusion. “Her leadership is a labor of love, and she is an inspiration to so many students on campus,” said Chloe Dolese Mandeville, senior assistant director for student engagement and access at the Allen School.
Joo Gyeong Kim was recognized for her foundational leadership in shaping the Allen School’s Changemakers in Computing (CIC) program that engages rising juniors and seniors in high school in learning about technology, society and justice. “As one of the founding mentors, she brought a thoughtful, mission-driven approach that helped define the program’s values and direction,” Dolese Mandeville said. She leaned heavily on that approach when she took on temporary leadership of the entire program one summer while both directors were out sick. Known as a steady and compassionate leader, Kim’s impact extends to the entire CIC community.
Anjali Singh was honored for her dedicated service in multiple roles with the Student Engagement & Access team. Starting with the Allen School Ambassadors — a team of current majors who engage middle and high school students in learning about computer science via school visits and field trips — Singh used her warmth and knack for storytelling to inspire students. She quickly rose to the position of lead ambassador before going on to help launch a new team of Student Recruitment Representatives. Having served hundreds of prospective students along the way, “her dedication, advocacy for accessible pathways into computing and long-standing service have left a lasting legacy,” Dolese Mandeville said.
Zhengyu Zhang was recognized for his service to the robotics research community in the Allen School. His contributions include the mastery of complex simulation tools, one-on-one mentorship and the development of an open-source repository that is used by researchers in multiple labs. Known for being generous with his time and willing to support others regardless of their skill level, Zhang’s collaboration, service and mentorship has, noted Dolese Mandeville, “enabled the success of countless students and researchers, from undergraduates to postdocs.”
Outstanding CSE Senior Awards
The epitome of scholarship and leadership: Balazinska with (left to right) Andre Ye, Kenneth Yang, Eujean Lee and Kianna Roces Bolante (Photo by Kerry Dahlen)
Balazinska called Bolante back to the stage to collect one of four awards designed to recognize students who demonstrate superior scholarship and leadership potential — qualities that Bolante has epitomized during her time at the Allen School. In addition to her aforementioned service contributions, she has also contributed to research supporting people with Parkinson’s disease, language preferences in disability communities and computer science education. For the latter, she developed a suite of six social computing modules which she piloted with more than 1,400 local high school students. Last year, she received a CRA Outstanding Undergraduate Researcher Award from the Computing Research Association — one of four in the nation — for her work.
Eujean Lee was recognized for her outstanding academic achievements and research contributions, for which she was also nominated for a College of Engineering Dean’s Medal for Academic Excellence. As an undergraduate researcher in the Makeability Lab, Lee co-authored two papers on the use of augmented reality and computer vision to make sports more accessible to people with low vision — one of which earned a Best Paper Award at the Workshop on Inclusion, Diversity, Equity, Accessibility, Transparency and Ethics in Extended Reality (IDEATExR). Lee also served as vice president of the Korean Job Search Association, helping to connect students with career opportunities and resources.
Honoree Kenneth Yang’s research spans software engineering, neuroscience and computer graphics. He contributed to a paper presenting a suite of new, more reliable version control merge tools for shared repositories such as Git that was published at the IEEE/ACM International Conference Automated Software Engineering (ASE) — one of the top conferences in the field. He also developed new software tools for automating electrophysiology experiments to accelerate brain research and open up new avenues of experimentation. Yang previously received a CRA Outstanding Undergraduate Researcher Award honorable mention for his work.
Andre Ye was recognized for blending technical innovation with humanistic insight in research that spans computer vision, machine learning and human-AI interaction. In the Allen School’s Social Futures Lab, he developed a framework to account for human uncertainty in medical image segmentation models that earned an honorable mention at the Conference on Human Computation and Crowdsourcing (HCOMP). He has also investigated the influence of linguistic and cultural differences on image captioning models and the use of language models to support critical thinking. Ye has earned multiple accolades, including a Paul & Daisy Soros Fellowship and a College of Arts & Sciences Dean’s Medal. He will pursue his Ph.D. at MIT in the fall.
Celebrating scholarly achievement
The path to a doctorate involves years of intensive, original research — as the 52 newly-hooded Ph.D. graduates seated on the floor of the arena could attest. But they are not the only Allen School students who make original contributions to the field on their way to earning a degree; a significant number of bachelor’s and fifth-year master’s students know their way around a lab, as well. Professors Leilani Battle and Maya Cakmak, co-chairs of the Allen School’s Undergraduate Research Committee, had the pleasure of highlighting several of them with Best Senior Thesis or Outstanding Master’s Thesis awards.
They know their way around a lab: Maya Cakmak (left) and Leilani Battle (right) with Sela Navot, Haoquan Fang, Andrew Shaw and Hayoung Jung (Photo by Matt Hagen)
Best Senior Thesis (Winner)
Winner Andrew Shaw was recognized for his thesis titled “Agonistic Image Generation: Unsettling the Hegemony of Intention,” which was the result of a collaboration with Outstanding Senior honoree Andre Ye. Under the guidance of Allen School professors Ranjay Krishna and Amy Zhang, Shaw combined computer science and philosophy to develop a novel image generation interface that actively engages users with competing visual interpretations of their prompts, in consideration of the sociopolitical context, to facilitate user reflection. The paper was accepted to the ACM Conference on Fairness, Accountability, and Transparency (FAccT).
Best Senior Thesis (Honorable Mention)
In his thesis titled “SAM2Act: Integrating A Visual Foundation Model with A Memory Architecture for Robotic Manipulation,” honorable mention recipient Haoquan Fang introduced new models that achieved state-of-the-art performance on existing benchmarks for robotic manipulation, plus a new benchmark for testing robots’ ability to act based on past information. Fung completed this work under the supervision of Allen School professor Dieter Fox and presented his results at the International Conference on Machine Learning (ICML) and the Conference on Computer Vision and Pattern Recognition (CVPR).
Outstanding Master’s Thesis (Winner)
Winner Hayoung Jung’s thesis, “Towards Inclusive Technologies: Examining Social Values and Harms in Large-Scale Sociotechnical Systems,” introduced technical approaches grounded in the social sciences to measure and mitigate human biases and social harms perpetuated by generative large language models and algorithmically driven platforms such as YouTube. Jung completed this work, which was published in multiple top-tier conferences, under the guidance of Tanu Mitra, a professor in the UW Information School and adjunct faculty member in the Allen School. Jung will begin his Ph.D. in computer science at Princeton University in the fall.
Outstanding Master’s Thesis (Honorable Mention)
In his thesis titled “On the Existential and Strong Unforgeability of Multi-Signatures in the Discrete Log Setting,” honorable mention recipient Sela Navot advanced new theories and protocols for generating secure digital signatures in distributed, multi-party scenarios such as blockchain systems. Navot completed this work, which was published at the International Conference on the Theory and Application of Cryptology and Information Security (Asiacrypt), under the guidance of Allen School professor Stefano Tessaro.
Honoring excellence in teaching
Game recognizes game: Undergraduate Teaching Award recipients Lauren Bricker (left) and Ruth Anderson (right) with Bandes Award winners Amal Jacob, Antonio Ballesteros and Naama Amiel (Photo by Matt Hagen)
Bob Bandes Memorial Awards
The Bob Bandes Memorial Award for Outstanding Teaching, which is named in honor of a graduate student who died in a skydiving accident in 1983, recognizes exceptional teaching assistants (TAs) who go above and beyond in service to the thousands of students who take Allen School courses each year. Over the past year, roughly 650 undergraduate or graduate students served as TAs; among those, nearly 250 individuals were nominated for Bandes Award recognition via over 600 nominations submitted by Allen School faculty and students.
Winner Naama Amiel served as a TA for CSE 351: The Hardware/Software Interface, no fewer than six times before her latest TA assignment with CSE 451: Introduction to Operating Systems. According to one nominator, “Anyone that talks to her can explain and help others struggling with the same things, so she creates a chain of learning that has impacts far beyond her conversations and office hours.”
Fellow winner Antonio Ballesteros was honored for his kindness and patience in meeting students where they are in his role as TA for two courses — once for CSE 391: System and Software Tools, and three times for CSE 331: Software Design and Implementation. “In every interaction with Antonio as a student, it is clear that he deeply cares about every student’s learning and experience in the course,” one nominator said.
The third and final winner, Amal Jacob, served as a TA for CSE 344: Introduction to Data Management a total of six times, for five different instructors. Known as patient, friendly, professional and dedicated, Jacob earned the appreciation of instructors for routinely picking up extra responsibilities — often before they even realized there were gaps that needed filling — and was heralded by at least one student nominator as “one of the best TA’s I have had in the Allen School.”
Friends and loved ones cheer for the graduates (Photo by Matt Hagen)
The Allen School also recognized three TAs with honorable mentions. Elizabeth Shirakian was a TA nine times for the Allen School’s revamped introductory programming course series, specifically CSE 121 and CSE 122, and will be the instructor for the summer offering of CSE 122. Megan Wangsawijaya was a TA multiple times for CSE 390T: Transfer Admit Seminar that helps newly arrived transfer students acclimate to the Allen School, as well as CSE 390Z: Mathematics for Computation Workshop, a companion to the Allen School’s Foundations of Computing course. Last but not least, Ph.D. student Zhihan Zhang earned “rave reviews” for his support of student teams enrolled in the CSE 475: Embedded Systems Capstone course.
Undergraduate Teaching Awards
Bolante presented the 2025 Undergraduate Teaching Awards in her capacity as chair of COM^2, the largest Allen School student organization that represents all undergraduate majors.
“As we celebrate the class of 2025, it’s worth remembering that none of us reached this stage alone,” Bolante said. “Educators do more than teach; they support us, inspire us and help shape the paths we take.”
Bolante described the first honoree, Ruth Anderson, as a “powerhouse” within the Allen School who has made a lasting impact on students as well as TAs. “In every class she teaches, Ruth creates a clear and supportive environment where students feel empowered to engage with complex material and build lasting understanding,” observed Bolante, noting that her work with TA’s elevates the quality of teaching across the school.
Anderson’s fellow honoree, Lauren Bricker, was Bolante’s first professor by way of the Allen School Scholars Program — making the presentation of this award especially meaningful. “Lauren brings warmth, enthusiasm and genuine care to absolutely everything she does…She creates inclusive spaces where students feel supported and encouraged to grow,” Bolante said. “Through her tireless support and advocacy, Lauren continues to inspire and uplift our community.”
