Christy Graves Email & Phone Number

Princeton, Nj

For my dissertation, I study mean field games, which is the continuum analog of game theory. In the usual setup in game theory, a game consists of some finite number of players. For games with a large number of players, solving for any Nash equilibria of the game is intractable, both theoretically and computationally. If we consider a large player game where all the players are symmetric, then we can approximate the game using mean field games. Instead of considering all of the many combinatorial possibilities of players’ actions, we can consider a single representative player responding to the continuous aggregate, i.e. the mean field, of the other players.My research in this field includes an application to jet lag, a study of numerical methods for solving such formulations, an investigation of the price of anarchy, and extending the theory to asymmetric players who are connected to one another via graphons. Each of these projects are detailed more on my website, with links to relevant papers and presentations.

Civis is a data science tech company based in Chicago, IL. One product that they offer is attribution for advertisements, which is the process of determining which ad creatives are the most effective. Civis applies their proprietary methods on ad exposure and store visit data to inform their clients about the relative effectiveness of their ad creatives. These proprietary methods are more sophisticated than the naïve industry standard method of last touch attribution, where a store visit is attributed to the most recent ad viewed by the customer. However, no ground truth exists to showcase the accuracy of their methods, and clients are not willing to risk losing profits by running experiments in the wild. The goal of this 10 week internship project was to generate synthetic data with known ad treatment effects and use this data to benchmark Civis' proprietary methods against the industry standards.This internship gave me experience working in an agile environment, with tasks organized with tickets, sprint planning, and story pointing with Jira. I also gained experience using SQL to query Redshift relational databases and containerization with Docker for computing in a cloud environment. Collaborating on a team gave me the opportunity to strengthen my knowledge of git version control and various Python libraries. This internship solidified my interest in pursuing data science as a career, and gave me the skills and experience necessary to be prepared for this career path.

Durham, Nc

My undergraduate thesis was the beginning of the Duke Quantifying Gerrymandering project, which has been used in the court case Common Cause v. Rucho. The goal of this thesis was to use outlier analysis to show that North Carolina's congressional districts used in the 2012 election were gerrymandered. This project involved gathering various datasets (vote, population, and geographic data), and implementing a Markov chain Monte Carlo algorithm to randomly sample districtings plans that divide North Carolina into 13 congressional districts. Finally, we retabulated election results to determine that the 2012 election was an extreme statistical outlier, and thus, gerrymandering was involved. Visit my website for links to relevant papers, presentations, and press.

Los Angeles, Ca

This was a blended internship/research experience through the RIPS (Research in Industrial Projects for Students) program at IPAM (Institute for Pure and Applied Mathematics) where my team was partnered with the Symantec Corporation. One service that Symantec offers is backup storage. Symantec needs to have enough capacity to store all of their clients' backups, allowing their clients to recover the state of their system at a previous point in time. However, Symantec did not have models to predict how much storage capacity would be needed to support all of their clients for future backups.The goal of the project was to give Symantec tools for predicting capacity usage in backup storage systems. Our approach was to first develop a detailed model of the backup process, including full backups, incremental backups, and the deduplication process. Next, we created probabilistic models of deduplication rates, client backup sizes, and models for forecasting future backup sizes. Finally, we used these models to produce Monte Carlo samples of future capacity usages, to arrive at a distribution of possible times that a given storage capacity could be reached. I presented our findings through a conference proceedings at the 2015 IEEE International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems. Visit my website for links to relevant papers and presentations.Note that Symantec split in 2015, with the division offering backup services becoming Veritas Technologies Corporation.

Doctor Of Philosophy - Phd, Applied & Computational Mathematics

Master'S Degree, Applied & Computational Mathematics

Bachelor Of Science (B.S.), Mathematics

Education record