Funded Projects
The Jurist Center provides research support for the following projects:
2023
- Graph Neural Networks for Software Security (Cristian Garces, Jiang Ming): Resolving indirect control flow is one of the fundamental challenges in software security analysis. Many analysis algorithms and security techniques rely on a precise indirect control flow result, such as recursive disassembling, control flow integrity, data-flow analysis, etc. This project explores how deep learning can be applied to indirect control flow resolving problems. The graph is a natural representation used in the program analysis domain. We utilize the graph neural network in our augmented control flow graph to learn how to predict indirect callees. We translate the indirect callee prediction problem into a graph's edge prediction problem. This work is under review for publication.
- Calibrating Deep Learning Models (Yunbei Zhang, Jihun Hamm): The student worked on the problem of calibrating deep learning models whose goal is to improve reliability of the predicted confidence output in addition to the usual performance measures such as accuracy. Ensuring good calibration of deep learning predictions is become more and more important for trustworthy applications as it allows one to make informed decisions based on the model output. Over the summer, the student developed novel techniques to improve calibration, evaluated the impact of the methods on different model architectures, and established best practices for implementing calibrated deep learning models across various application domains.
- publication: A Mehra, Y Zhang, J Hamm (2023) "Analysis of Task Transferability in Large Pre-trained Classifiers," Workshop on Mathematics of Modern Machine Learning (M3L) at NeurIPS 2023.
- publication: Mehra, A., Zhang, Y., Kailkhura, B., & Hamm, J. (2024). On the Fly Neural Style Smoothing for Risk-Averse Domain Generalization. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (pp. 3800-3811).
- Making Human-Like Moral Decisions (Disa Sariola, Nicholas Mattei): This project conducts an analysis of human subjects data in order to guide automated decision making in diverse environments. The goal is to develop AI agents that can mimic human behavior in these ethically constrained decision environments, with a long term research goal to use AI to help humans in making better moral judgments and actions.
- Forecasting Vaccination Decisions using Social Media (Xintian Li, Aron Culotta): The goal of this project is to understand how COVID-19 vaccination decisions are made by analyzing social media data. Xintian’s work develops natural language processing methods to extract personal expressions of vaccination intent from social media, and to develop forecasting models to predict how evacuation intent will evolve during the course of the pandemic.
- Improving the Efficiency of Large Language Models (Sofiia Druchyna, Lu Peng): The goal of this project is to analyze recent LLM models running on 1-2 GPUs from an architectural perspective. The project will analyze the tradeoff among resource usage, power consumption, and model accuracy. A potential approximate model can be proposed to simplify the LLM models.
2022
- Computational Epitome Prediction (Avik Bhattacharya, Ramgopal Mettu, Sam Landry): This project develops data-driven models to predict cancer mutations to target for therapy. The Jurist support for Summer 2022 enabled large scale data collection related to bladder cancer, and the implementation of predictive models.
- Understanding Emerging Issues in Public Schools from Online Reviews (Linsen Li, Aron Culotta, Nicholas Mattei): The goal of this project is to determine whether the language used in a school’s online reviews is a leading indicator of future changes at that school. For example, can we predict future changes in a school's demographics or test results based on how people describe the school online? Doing so has the potential to help us better understand how families select schools, what attributes are important to them, and how inequities may be exacerbated by online mechanisms. The work is in collaboration with Dr. Douglas Harris (Economics). The Jurist support for Summer 2022 enabled the collection and analysis of millions of online reviews from over 60K schools in the U.S.
- Robust Reinforcement Learning for Security (Xiaolin Sun, Zizhan Zheng, Nicholas Mattei): This project investigates how randomized smoothing can help obtain policies with certified adversarial robustness for (deep) reinforcement learning tasks subject to state perturbation attacks. The goal is to leverage state-of-the-art RL algorithms to learn smoothed policies more effectively, providing insights into deploying RL in security and safety-sensitive domains. The Jurist support for Summer 2022 enabled researchers to implement new RL algorithms and evaluate them on security domain benchmarks.
- publication: Xiaolin Sun, Jacob Masur, Ben Abramowitz, Nicholas Mattei, Zizhan Zheng (2023) "Pandering in a (Flexible) Representative Democracy," Proceedings of the Thirty-Ninth Conference on Uncertainty in Artificial Intelligence.
- publication: Sun X, Masur J, Abramowitz B, Mattei N, Zheng Z. (2023) Does Delegating Votes Protect Against Pandering Candidates?. In Proceedings of the International Conference on Autonomous Agents and Multiagent Systems.
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:
- Learning to Secure Cooperative Multi-Agent Learning Systems: Advanced Attacks and Robust Defenses, Zizhan Zheng
- Fair Recommendation Through Social Choice, Nicholas Mattei
- Mechanisms and Algorithms for Improving Peer Selection, Nicholas Mattei
- Modeling and Learning Ethical Principles for Embedding into Group Decision Support Systems, Nicholas Mattei
- Machine Learning for Advanced Manufacturing, through the Louisiana Materials Design Alliance, Jihun Hamm
- Quantifying Morphologic Phenotypes in Prostate Cancer - Developing Topological Descriptors for Machine Learning Algorithms, Carola Wenk, J. Quincy Brown, Brian Summa
- Scalable, Content-Based, Domain-Agnostic Search of Scientific Data through Concise Topological Representations, Brian Summa
- Scalable Interactive Image Segmentation through Hierarchical, Query-Driven Processing, Brian Summa
- Predicting Real-time Population Behavior during Hurricanes Synthesizing Data from Transportation Systems and Social Media, Aron Culotta
- Quantifying Multifaceted Perception Dynamics in Online Social Networks, Aron Culotta
- Understanding the Relationship between Algorithmic Transparency and Filter Bubbles in Online Media, Aron Culotta
- Reducing Classifier Bias in Social Media Studies of Public Health, Aron Culotta
