Events

Website

Project description: Independent redistricting commissions and courts are increasingly relying on algorithms to assess whether a given map is gerrymandered (meaning intentionally drawn to disadvantage a political party or group). The dominant, data-driven approach relies on sampling a massive ensemble of “random” maps, formally represented as partitions of a graph, where the vertices represent indivisible geographic units like census blocks or voting precincts. This is a challenging computational problem involving up to hundreds of thousands of vertices. We have developed a new, theoretically faster algorithm leveraging additional topological information about the embedding of the underlying graph in the plane, structural data that existing algorithms do not take into account, and is not currently available in any data repositories (but should be able to be inferred from the geographic data). We would like to implement this algorithm and contribute to the open-source GerryChain Python package. The ideal candidate would be someone with exceptional programming skills and an ability to reason about difficult geometric/topological problems.

Website

Website

Website

Project Description: The binary perceptron is a deceptively simple model at the crossroads of machine learning theory, statistical physics, and computational complexity. It asks: can we find a vector of {1, -1}’s that satisfies a large number of random constraints? The mystery is that solutions live in rare, “frozen” regions of the search space, seemingly inaccessible to most algorithms, yet somehow discoverable by certain ingenious methods. In this project, we will study these frozen structures, analyze the behavior of different classes of algorithms on them, and investigate the “information–computation gap,” where problems are theoretically solvable but computationally challenging. We will also explore the perceptron’s connection to real-world tasks in causal inference, where balancing covariates and designing experiments are closely related to perceptron-style constraints. Students will gain hands-on experience in probability,  algorithms, computational experiments, and causal inference, while engaging with questions that connect deep theory to impactful applications.

Website

Website

Project Description: Capitalizing on the widespread success of machine learning (ML) models in predictive tasks, the field of learning-augmented algorithms seeks to answer a key question: Can we enhance the performance of classical algorithms by incorporating the predictions of ML models trained on historical data? A critical challenge in this context is that the accuracy of these predictions is unknown. Therefore, learning-augmented algorithms are designed with two main objectives: achieving optimal performance when predictions are accurate and maintaining the worst-case guarantees of classical algorithms when predictions are arbitrarily poor. The goal of this project is to leverage recent theoretical developments in the area in real-world applications.

Website.

Project Description: Neighborhood conditions are powerful predictors of health, education, and economic opportunity, yet widely used measures like the Area Deprivation Index (ADI) are based on survey data updated infrequently. This project uses Google Street View imagery, satellite data, and U.S. Census indicators to train machine learning models that quantify neighborhood disinvestment at fine geographic scales.

Students will help develop computer vision and geospatial models, engineer features from imagery, and test model performance against existing indices. They will also explore bias detection and validation using real-world health and education outcomes. The goal is to produce a more timely, precise, and equitable measure of neighborhood disadvantage that can inform policy and research.

Website.

Project Description: The Gerstein lab conducts bioinformatics/data science research in the biomedical and genomic fields. We use various computational analytics methods including machine learning and AI techniques to analyze large biomedical datasets. The lab has particular focuses on the following areas of research: genomic privacy, personal genomes, genome annotation and neurogenomics.

Mark Gerstein is the Albert L. Williams Professor of Biomedical Informatics, Molecular Biophysics & Biochemistry, Computer Science and Statistics and Data Science.

Website.

Project Description: Gastric cancer remains a leading cause of cancer mortality, and CLDN18.2 is an emerging therapeutic target. Prioritizing antibody candidates for follow-up experiments is still slow and labor-intensive. Our lab has developed a deep-learning framework for antibody structural modeling that represents antibody–antigen interfaces as graphs and learns to rank pairs by relative affinity from 3D geometric and contact features. In this project, we will adapt that framework to gastric cancer and benchmark it against alternatives that rely only on sequence information, as well as current structure-prediction and generative approaches (described methodologically, without naming specific models).

Methodology: The student will (1) reproduce our baseline structure-aware ranking pipeline on a curated antibody–antigen dataset, verifying quality metrics and generating sanity-check plots; (2) apply the pipeline to CLDN18.2 by first building structural representations of the target with state-of-the-art prediction tools, assessing confidence with standard quality metrics, and, if needed, refining with geometric and energy-based procedures; (3) generate a candidate list using established sequence-based methods for antibody–antigen interaction scoring; and (4) construct structural models for the top candidates and apply our structure-aware ranking framework to prioritize antibodies for experimental validation. All code and results will be delivered as reproducible notebooks with clear evaluation tables and ablation notes.

Conclusion: The project will deliver a tested, end-to-end pipeline and a prioritized shortlist of CLDN18.2 antibody candidates, along with quantitative evidence of when structure adds value beyond sequence, directly advancing our gastric-cancer goals.

Desired skills: Python, machine learning, familiarity with deep learning for protein modeling (or willingness to learn), data wrangling, and Git; interest in immunology/structural biology is a plus.

Website.

Project Description: This project explores how students can better learn probability—and quantitative subjects more generally—outside the classroom, where most learning is believed to occur. Building on tools such as QR-linked course booklets, semi-automated feedback and grading, two-stage homework, solution videos, and gamified practice problems, the goal is to investigate how such resources can be designed and integrated to improve learning outcomes.

Project Description: AI has mastered structured games like chess, Go, Diplomacy, and poker. But real-world bargaining—central to commerce—remains unsolved. In negotiations, buyers conceal willingness to pay and sellers hide costs. These values are not only hidden from the bargaining opponent but also from researchers trying to design negotiation agents, creating a fundamental challenge that distinguishes this setting from AI for board games.

Our interdisciplinary group is developing AI negotiators: large-language-model agents that learn to bargain under these conditions. We combine economic theory with modern machine learning, and we are collaborating with eBay, where millions of peer-to-peer negotiations provide a unique testing ground. Projects include analyzing negotiation transcripts from eBay data, developing reinforcement learning methods to train bargaining agents under hidden information, and evaluating performance in terms of trade frequency, gains from trade, and strategic behavior.

Students will gain experience in reinforcement learning, LLM fine-tuning, and empirical analysis of bargaining data. This is an opportunity to contribute at the intersection of economics, AI, and real-world marketplaces.

Project Description: With the increasing threats from climate change, cities need to rethink how they use space to protect people from rising environmental risks. This project focuses on identifying and assessing the potential of underutilized gray spaces, such as parking lots, to be transformed into climate-resilient areas. By applying computer vision techniques to satellite and street-view imagery from cities across the globe, the goal is to map these gray spaces, assess their potential for transformation, and quantify the resulting benefits, such as reduced urban heat and flood risks.

Students will gain hands-on experience in spatial data analysis, computer vision, and machine learning while contributing to solutions for climate resilience and sustainable urban development.

Requisite Skills and Qualifications: Ideal candidates should have a good understanding of GIS and be proficient in handling large datasets. Experience with Python and computer vision techniques would be beneficial.

Website.