[CS] Ted Shao Dissertation Defense/May 28, 2025

[via cs]

This is an announcement of Ted Shao's Dissertation Defense.
===============================================
Candidate: Ted Shao

Date: Wednesday, May 28, 2025

Time:  4 pm CST

Remote Location: https://urldefense.com/v3/__https://uchicago.zoom.us/j/99755524175?pwd=EhyHZeRjNaY46eokPaEHcHw3l2sg6L.1&jst=2__;!!BpyFHLRN4TMTrA!_QS0KCXcullMpnxWW56TLC8Kk8rnw8Dr7Z7-I6dpiuDL99KHE8-8RvJAmRxm48dHeJMtSHzQSRhyKaKC7LY$

Location: JCL 298

Title: Low-Latency, Private, and Observable Model Serving: A Data-CentricApproach for Distributed Streams

Abstract: The proliferation of Compound AI systems, whichintegrate multiple models and data streams for real-time decision-making,presents significant challenges in achieving low operational latency,maintaining system observability, and ensuring user privacy. Addressing theserequires a fundamental shift from optimizing machine learning models or systemcomponents in isolation to a holistic approach. This thesis exploresdata-centric co-design principles that optimize the interplay between ML requirementsand data management strategies. This work contributes: (1) Real-time modelrouting strategies for incoming streaming data to reduce latency whilepreserving accuracy. (2) A decentralized observability framework thatsignificantly minimizes logging overhead and respects data borders. (3)Algorithmic data minimization techniques to effectively protect useridentifiability while preserving model utility.

Advisors: Sanjay Krishnan

Committee: Michael Franklin, Nick Feamster, Sanjay Krishnan