CSE REU Projects for 2026
Game Theoretic Analysis of American Football
Faculty: Yevgeniy Vorobeychik
This project will aim to develop a principled game-theoretic model of American football. At the core of the project will be a predictive model of the distribution of yards gained by the offense as a function of the formation and play call for the offense and defense. To this end, we will use the NFL big data bowl data from Kaggle. Next, we will develop a game-theoretic model in which the offense aims to maximize the expected number of points scored on the current drive, which the defense aims to minimize. Finally, the project will develop solution approaches which prescribe joint formation and play selection strategies for both the offense and defense.
Explainable Planning and Scheduling
Faculty: William Yeoh
In human-aware planning and scheduling systems, when the agent recommends a plan or schedule to a human user, it is often the case that the user might not understand why the recommendation is good, for example, compared to an alternative in the user’s mind. In such a scenario, there is a need for the agent to explain its recommendation to the user, providing them with the necessary information to understand properties of the recommendation (e.g., optimality, feasibility, etc.).
In this REU project, students will have the opportunity to investigate solution approaches from a wide spectrum, ranging from symbolic logic-based approaches that use knowledge representation and reasoning (KR) to data-driven approaches that use large language models (LLMs), as well as neuro-symbolic approaches that combine the benefit of both.
Skills Required: Strong programming skills. Familiarity with logic and/or LLMs is a plus.
Post-Training for LLMs
Faculty: Jiaxin Huang
The project seeks to improve LLM post-training in various settings and directions. For example, we can explore RLVR for open-ended tasks, the interplay between various training paradigms (SFT and RL) or reward signals (outcome rewards and process rewards), as well as synthesize dynamically challenging questions for training. We can also explore applying RL techniques to agentic settings.
Skills Required: Strong programming skills, prior experience with research projects on deep learning is highly desirable.
Neurosymbolic AI and linear codes
Faculty: Netanel Raviv
Neurosymbolic AI is an approach that combines neural networks with symbolic AI systems. This integration aims to create AI systems that have both the learning capabilities of neural networks and the reasoning capabilities of symbolic methods. Neurosymbolic AI recently featured as the first among six possible futures for AI research in a white paper by the Community Consortium of the Computing Research Association. Vector Symbolic Architectures (VSAs) are a popular approach for designing neurosymbolic systems, in which atomic objects are represented using long quasiorthogonal vectors, that are manipulated algebraically to create compositional representations. Recently, linear error correcting codes have been shown to be suitable for several neurosymbolic tasks, such as encoding and recovery. This project will explore this connection further by developing efficient algorithms, code constructions, and bounds.
Skills Required: An ideal candidate will have strong mathematical skills, especially in linear algebra, finite fields, and probability. The student must be comfortable with reading and writing proofs, have critical and algorithmic thinking, and also be capable of implementing simple learning algorithms in Python. Familiarity with error correcting codes or associative memories (Hopfield networks) is appreciated.
Guiding exploration in benign environments
Faculty: Brendan Juba
It is challenging for an agent to learn a useful model of its environment by trial-and-error. Even seemingly simple examples of standard environment models capture problems like combination locks, where there are no efficient strategies for reliably discovering key features of the environment, such as opening the lock. Recent works have proposed information-theoretic quantities that capture the inherent difficulty of such exploration problems: these measures capture when it is possible in principle to learn an adequate model of the environment in a small number of moves. In this project, we aim to study computational aspects of the problem. We are seeking algorithms that can solve nontrivial families of such exploration problems.
Skills Required: Participants should be comfortable with reasoning about probability and analyzing algorithms. Familiarity with classical AI planning and/or machine learning theory would be helpful.
High-Performance Computing to Benefit Astrophysical Observation
Faculty: Roger Chamberlain, Jeremy Buhler
Astrophysics tries to understand the physics of the universe by observing energetic phenomena — such as supernovae, neutron star mergers, and active galactic nuclei — in deep space. The instruments that observe these phenomena include telescopes in all bands of the electromagnetic spectrum, as well as specialized sensors for other signals such as gravity waves and neutrinos. Not surprisingly, making sense of all this data requires advances in computation. This is particularly true when the phenomena are transient (appearing and disappearing within seconds to minutes); multiple telescopes must coordinate to detect and computationally analyze volumes of data from such transients in real time in order to maximize the gain in scientific knowledge.
In this project, we work with astrophysics experts to improve the performance and power efficiency of computations that benefit scientific observation. Because these computations will ideally happen aboard a telescope in space or a balloon in Earth’s upper atmosphere, they must meet strict hardware size, weight, and power (SWaP) constraints as well as achieving goals for application throughput and latency. To succeed within these constraints, we turn to low-power multicores and to non-traditional computer architectures (FPGAs, GPUs) to accelerate an application. Students will have the opportunity to work on computations for real astrophysical missions such as the Advanced Particle-astrophysics Telescope (APT) and the Compton Spectrometer and Imager (COSI).
Skills Required: Programming ability; Python; C/C++ experience a plus but not strictly required
AI For Health
Faculty: Chenyang Lu
Join the AI for Health Institute this summer to apply cutting-edge AI to real-world challenges in health care and public health. You will work closely with graduate students, faculty, and clinicians on projects that aim to improve human health at scale.
Example research topics include: Predicting clinical outcomes of surgeries and mental health interventions; 2) Designing safe and responsible AI chatbots to support mental health care; 3) AI-powered cancer screening using blood and tissue samples.
Students will gain hands-on experience with modern AI methods, real clinical and public-health data, and interdisciplinary research at the intersection of computer science and medicine.
Skills Required: Prior coursework or experience in machine learning, artificial intelligence, computer vision, or large language models.