Data Science Institute, The University of Chicago

Please join us for our monthly Chicago Data Night, cohosted by ChiData and the DSI, and this month with special partner, Booth’s Center for Applied AI. Practitioners, academics, and aficionados within the Chicago area are all invited to be part of a community at the intersection of industry and academia, brought together by a mutual interest in data. Each month, guest speakers will cover a specific data-related topic.

Hors d’oeuvres and drinks will be provided. Admission is free, and we strongly encourage an RSVP to attend.

NEW HYDE PARK MEETING LOCATION!
Data Science Institute, Faraco Family Commons 105
5460 S University Avenue
Chicago, IL 60615

AGENDA
5:30pm: Networking Reception
6:00pm: Welcome Remarks
6:05pm: Guest Speaker
6:50pm: Q&A
7:00pm: Networking Reception
7:30pm: Event Concludes

Title: Federation Over Text: Insight Sharing for Multi-Agent Reasoning

Abstract: LLM-powered agents often reason from scratch when presented with new problem instances and can lack automatic mechanisms to transfer learned skills to other agents. In this talk, I discuss a distributed inference framework, Federation over Text (FoT), that enables multiple clients solving different tasks to collectively generate a shared library of metacognitive insights by iteratively federating their local reasoning processes. Instead of federation over gradients (e.g., as in distributed training), FoT operates at the semantic level without any gradient optimization or supervision signal. Iteratively, each client runs an LLM agent that does local thinking and self-improvement on their specific tasks independently, and shares reasoning traces with a central server, which aggregates and distills them into a cross-task (and cross-domain) insight library that existing and future agents can leverage. I present a set of applications show that FoT improves reasoning effectiveness and efficiency across a range of applications, including mathematical problem solving, cross-domain collaboration, real-world daily tasks, and machine learning research insight discovery.

Bio: Tian Li is an Assistant Professor at the Computer Science Department and Data Science Institute at the University of Chicago. She is also a member of the UChicago Committee on Computational and Applied Mathematics. Her research interests are in optimization, trustworthy machine learning, and collaborative learning. She has spent time at Microsoft Research Asia, Google Research, and Meta Foundational AI Research Labs. Tian’s work has won several awards, including Google Research Award, Signal Processing Magazine Best Paper Award, and the first place in the Privacy-Enhancing Technologies Challenge. She received her PhD in Computer Science from Carnegie Mellon University and BS degrees in Computer Science and Economics from Peking University.