BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Foundation Models for Structured Data - Xiyuan Zhang\, Amazon Web 
 Services
DTSTART:20260127T160000Z
DTEND:20260127T170000Z
UID:TALK236053@talks.cam.ac.uk
CONTACT:Cecilia Mascolo
DESCRIPTION:Foundation models are transforming structured data learning mu
 ch like large language models did for text. In this talk\, I will present 
 our new foundation models\, Mitra and Chronos-2\, which demonstrate how sy
 nthetic pretraining and in-context learning (ICL) enable models to general
 ize across diverse tabular and time-series tasks without task-specific tra
 ining. Mitra curates a principled mixture of synthetic priors to achieve s
 tate-of-the-art performance on tabular classification and regression\, whi
 le Chronos-2 introduces group attention to unify univariate\, multivariate
 \, and covariate-informed forecasting. Together\, they illustrate a new pa
 radigm where the design of synthetic data priors and ICL mechanisms\, rath
 er than per-task fine-tuning\, drives generalization and scalability acros
 s structured domains.\n\nBio: Xiyuan Zhang is an Applied Scientist at Amaz
 on Web Services working on machine learning for structured data (time seri
 es\, tabular)\, especially on pre-training and multimodal analysis. She is
  the lead author of Mitra\, the most downloaded tabular foundation model o
 n HuggingFace\, and co-author of Chronos\, the most downloaded time series
  foundation model on HuggingFace. Xiyuan earned her PhD in Computer Scienc
 e from the University of California\, San Diego. She is a recipient of the
  Qualcomm Innovation Fellowship and has been recognized as a Cyber-Physica
 l-System (CPS) Rising Star.
LOCATION:Online
END:VEVENT
END:VCALENDAR