BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:LLM Processes for Regression and Classification - John Bronskill (
 University of Cambridge)
DTSTART:20250409T150000Z
DTEND:20250409T153000Z
UID:TALK229900@talks.cam.ac.uk
CONTACT:Cat Spencer
DESCRIPTION:Machine learning practitioners often face significant challeng
 es in formally integrating their prior knowledge and beliefs into predicti
 ve models. Our goal is to build prediction models that can process numeric
 al data and make probabilistic predictions\, guided by natural language te
 xt which describes a user’s prior knowledge. Large Language Models (LLMs
 ) provide a useful starting point for designing such a tool since they pro
 ve 1) an interface where users can incorporate expert insights in natural 
 language and 2) an opportunity for leveraging latent problem-relevant know
 ledge encoded in LLMs that users may not have themselves. We show how LLMs
  can compute joint posterior predictive distributions over an arbitrary nu
 mber of outputs that may be numeric or categorical in settings such as tim
 e series forecasting\, multi-dimensional regression\, black-box optimizati
 on\, image modeling\, and tabular data. Finally\, we demonstrate the abili
 ty to usefully incorporate text into numerical predictions\, showing how t
 he text influences the predictive distribution and improves predictive per
 formance.\n \nReferences:\nLLMProcesses: Numerical Predictive Distribution
 s Conditioned on Natural Language https://arxiv.org/pdf/2405.12856\nJoLT: 
 Joint Probabilistic Predictions on Tabular Data Using LLMs https://arxiv.o
 rg/pdf/2502.11877
LOCATION: Cambridge University Engineering Department\, CBL Seminar room B
 E4-38.  For directions see http://learning.eng.cam.ac.uk/Public/Directions
END:VEVENT
END:VCALENDAR