Event Details

PhD Thesis Proposal - Siddartha Devic

Thu, Oct 30, 2025
12:15 PM - 1:45 PM
Location: GCS 502c
Title: Fonudations of Trustworthy Machine Learning in Modern Algortihmic Systems
 
Date and Time: Thursday, October 30th, 2025 - 12:15p
 
Location: GCS 502C
 
Committee Members: Vatsal Sharan, David Kempe, Shanghua Teng, Angela Zhou, Aleksandra Korolova
 
Abstract: Modern algorithmic platforms like LinkedIn, ride-sharing, content delivery, and chatbots are increasingly complex aggregations of multiple machine-learned systems. This thesis seeks to investigate and understand how the foundations of trustworthy and reliable machine learning can be framed and studied within some of these complex systems. We begin by investigating the usefulness of unlabeled data for foundational supervised learning tasks, and how it relates to classic concepts like regularization. Next, we seek to understand how fairness and reliability can be understood or be achieved in complex algorithmic marketplaces such as ranking and matching systems. Finally, we investigate foundations of model uncertainty quantification in a variety of modern settings such as supervised learning over diverse non-homogenous populations, large language models, and model routing. Together, the work presented in this thesis helps advance the understanding, reliability, and fairness of modern machine learning systems and platforms.