SSE Comp Sci Dept.

Projects

Funded Projects

The Jurist Center provides research support for the following projects:

2026

  • Risk-Conditioned RLHF and Rare Event Estimation in LLM Policies (Zixuan Liu, Zizhan Zheng):

    This project focuses on reinforcement learning and AI alignment along two directions. First, it will study risk-conditioned fine-tuning of large language models, aiming to develop a reinforcement learning from human feedback (RLHF) algorithm that can adapt to a desired risk level specified at inference time. Second, it will explore new sampling and white-box methods for estimating rare events in LLM policies, with particular emphasis on challenging multi-turn decision-making settings.

  • Evaluating Self-Evolving AI Agents (Yilun Zhou, Jihun Hamm):

    This project will investigate how to evaluate self-evolving AI agents—specifically, how to measure the quality of evolution itself rather than only end-state task performance. The field has produced dozens of self-evolving methods, but there is no shared definition of what "good evolution" means, leaving properties like retention, efficiency, and robustness inconsistently measured. The project will formalize a small set of evaluation axes grounded in recent surveys and apply them to representative methods to surface differences hidden by current benchmarks. The expected outcome is an evaluation framework and a draft paper for publication.

  • Intelligent Visual Interactions for Model Steering in Oceanography (Yan Zhu, Rebecca Faust):

    This project will investigate intelligent interactions for model steering, with an application in the domain of oceanography via coral band identification. The goal is to design visual interactions that elicit human feedback to integrate into an underlying AI model. For this application, the project will study how feedback can be elicited via visual interactions over CT images of coral skeletons and integrated into an underlying segmentation model to improve AI predictions of banding patterns, both locally (for a specific user) and globally (across multiple users).

  • Continual Learning for Anomaly Detection in Human Movement Data (Chanuka Algama, Carola Wenk):

    This project aims to develop a continual learning framework for anomaly detection in human movement data that can adapt as normal mobility behavior changes over time. Movement data captures evolving patterns of life, but it is noisy, temporally dependent, spatially constrained, and often poorly aligned with the stationary assumptions of standard machine-learning models. While transformer-based sequential models have shown promise for learning trajectory patterns, they are typically evaluated under controlled conditions that do not reflect real-world non-stationarity. The project will address that limitation by designing methods that update their understanding of normal behavior while preserving sensitivity to meaningful deviations from expected patterns of life. These methods will be developed and evaluated using a cutting-edge large-scale simulated movement dataset developed through prior work and slated for public release, whose naturally emerging patterns of life provide a concrete testbed for studying anomaly detection under evolving behavioral conditions.

    • Publication: C. Algama “Drift or Deviation? Continual Detection of Anomalies Amid Changing Patterns of Life”, accepted to ACM SIGSPATIAL PhD and Graduate Studies Workshop, 2026.
  • Systematic Testing Methods for AI Agents (Ruchira Manke, Hridesh Rajan):

    This project will design systematic testing methods for AI agents that use large language models and external tools to solve complex tasks autonomously. As these agents operate in dynamic environments, testing them is challenging. The goal is to evaluate their reasoning behavior, tool use, and robustness under different operational conditions with the aim of building more reliable and trustworthy AI agents.

  • Designing AI-Based Reading Support for Deaf Users (Nazmum Khanom, Saad Hassan):

    This project will investigate design parameters for AI-based reading support technologies for Deaf users, including both text-based methods and sign language translation. Interface choices shape users' comprehension, trust, and efficiency, and the design space is defined by parameters such as how input text is marked and segmented, where and how support is presented (e.g., in-place, overlays, or external interfaces), and how long it remains visible. Additional parameters include how outputs, whether simplified text or sign language representations, are signaled and linked to the original content, especially given the lack of one-to-one correspondence between source and support. For sign language-based systems, further design dimensions such as video segmentation, navigation between text and video, and signer representation introduce added complexity. The project will explore these parameters through rapid prototyping and evaluations with DHH readers to identify optimal interface configurations.

  • LLM Game Theory (Harper Lyon (+ Disa), Nicholas Mattei):

    This project investigates how LLM agents play games with various levels of information, including formal reasoning capabilities. The work extends ongoing research in LLM game theory and includes opportunities for undergraduate researchers to contribute.

2025

  • Personalized Deep Learning with functional MRI (Zixiang Yin, Zhengming Ding):

    Deep learning has revolutionized computer vision, yet annotation inconsistencies limit its reliability in complex tasks like autonomous driving and medical diagnosis. To overcome these limitations, we will explore functional MRI (fMRI) as a means to personalize deep learning models by capturing individual perceptual and cognitive patterns. We plan to develop a dual-decoding framework designed for both biometric and semantic interpretation of neural data. By integrating deep learning with neuroimaging, this project enhances neural interpretability and advances our understanding of brain-based visual processing.

  • Blockchain Optimization using Deep Learning (Md. Ahsan Habib, Lu Peng):

    This project will investigate deep learning to optimize blockchain resource optimization and division.

    • Publication: M. A. Habib* and L. Peng, “From Topology to Performance: Workload-Driven Scalability Limits in the NEAR Sharded Blockchain,” Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics (SMC), Bellevue, WA, Oct. 2026.
    • Publication: M. A. Habib* and L. Peng, “Regular-Anatomy of NEAR Sharded Blockchain: A Longitudinal Workload Characterization,” Proceedings of the IEEE International Symposium on Workload Characterization (IISWC), Boulder, CO, Sep. 2026.
  • Explainable Machine Learning for Immunotherapy (Jiarui Li, Ramgopal Mettu):

    Understanding the mechanisms of T cell receptor (TCR) binding is crucial for developing immunotherapies against cancer and viruses. The proposed research will continue development of explainable machine learning techniques to uncover potential binding mechanisms.

  • Improving Urban Accessibility with Participatory AI (Syeda Mah Noor Asad, Saad Hassan):

    This project will explore how community groups collect urban accessibility data to inform the design of improved reporting and tracking tools. We will begin with formative interviews to understand the experiences, preferences, and challenges faced by both individual citizens and civic organizations involved in reporting and managing this data. Building on these insights, we will conduct participatory design sessions, introducing stakeholders to existing and emerging technologies—including AI and visualization tools—to co-design more effective systems. Outcomes will include design guidelines and two prototypes of reporting and tracking interfaces.

2024

2023

2022


Faculty Profiles

Our faculty lead projects across a broad array of artificial intelligence topics, including machine learning, multi-agent systems, computer vision, natural language processing, visualization, and ethics. Below is a sample of such projects: