Kweku Kwegyir-Aggrey | Brown University

About Me

I am a PhD Candidate at Brown University advised by Seny Kamara and Suresh Venkatasubramanian, and an incoming intern at IBM Research. I’ve also held positions as a research intern at Adobe and Arthur AI. Before graduate school, I studied Computer Science and Mathematics at the University of Maryland - College Park, where I worked with John Dickerson.

My research investigates foundational questions about uncertainty pertaining to responsible machine learning. Most often, I design and analyze algorithms that provably identify and mitigate issues surrounding data-driven decisions. Specifically, I use statistical techniques to examine the various sources of uncertainty which can obscure the responsible deployment of automated tools in real-world contexts. The high level goal of my research is to produce new ideas that can provide insight into existing and emerging challenges at the intersection of society and computation.

Research Interests

Machine Learning, Optimal Transport

Conference Publications

The Misuse of AUC: What High Impact Risk Assessment Gets Wrong
Kweku Kwegyir-Aggrey, Marissa Gerchick, Malika Mohan, Aaron Horowitz, Suresh Venkatasubramanian. Proceedings of ACM Conference on Fairness, Accountability, and Transparency (FAccT ‘23) [arxiv].

Workshop Publications

Model Selection’s Disparate Impact in Real-World Deep Learning Applications
Jessica Zosa Forde, A. Feder Cooper, Kweku Kwegyir-Aggrey, Christopher De Sa, Michael Littman. Science of Deep Learning Workshop (ICLR ‘21) [PDF].
Geometric Repair for Fair Classification at Any Decision Threshold
Kweku Kwegyir-Aggrey, Jessica Dai, A. Feder Cooper, Keegan Hines. The Workshop on Artificial Intelligence for Social Good (AAAI ‘23) [PDF].

Manuscripts

Everything is Relative: Understanding Fairness with Optimal Transport
Kweku Kwegyir-Aggrey, Rebecca Santorella, Sarah M. Brown [PDF].

Ongoing Projects

Measuring Fairness when Protected Groups are Unkown
with Naveen Durvasula and John Dickerson
Accessing and characterizing the accuracy of fairness metrics, in the common scenario where protected attributes are difficult or impossible to obtain.