Liam O'Carroll

Home Institution
Northwestern University

Year Participated
2022

Year in School
Undergraduate

REU Faculty Mentor
Matus Telgarsky

Research Area Interest
Theory and Algorithms

Project Title
Stochastic linear optimization with Markovian data

Biography & Research Abstract

Abstract:

Classical theoretical analyses of machine learning algorithms require the training data to be sampled independently and identically, but this does not coincide with real-world data which is highly dependent. Recently there has been significant progress in building theory for more general data, but much is still to be done. The goal of our research project is to prove stronger guarantees for learning a linear classifier using stochastic gradient descent (SGD) under data from a Markov process (a very popular way to model data dependencies).

Bio:

I'm a rising senior at Northwestern university majoring in computer science. I'm broadly interested in theoretical computer science, particularly subfields involving analysis and probability theory, including optimization theory, machine learning theory, and deep learning theory. In the past, I've done research in non-convex optimization, particularly rank-constrained semidefinite programming.