BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:PolymathicAI: Scaling up a Generalist AI Model for Science - Dr Mi
 les Cranmer (DAMTP/Physics/IoA)
DTSTART:20241031T140000Z
DTEND:20241031T150000Z
UID:TALK223909@talks.cam.ac.uk
CONTACT:Sri Aitken
DESCRIPTION:Machine learning is data hungry. Consider the massive contrast
  between general relativity\, which is still making accurate predictions a
 bout the extreme behavior of black holes over a century after it was origi
 nally proposed\, and the poor out-of-distribution predictions of a deep ne
 ural network trained from scratch on a small dataset. While machine learni
 ng is having a fantastic year in the physical sciences\, picking up two No
 bel prizes\, many of the scientific problems we wish to solve have very li
 mited training data available. \n\nNow\, the boom of Large Language Models
  from 2022 onwards has demonstrated the power of increased scale\, more ge
 neral models\, and more diverse pretraining. Motivated by this ongoing con
 firmation of Sutton's bitter lesson\, PolymathicAI\, a research collaborat
 ion mostly split between the University of Cambridge and Flatiron Institut
 e\, aims to build industry-scale foundation models for scientific tasks. I
 n particular\, our foundation models are specifically pretrained to work o
 n numerical datasets\, rather than language. I will present this initiativ
 e\, describe our ongoing work and upcoming releases\, and give some genera
 l comments about foundation model-based machine learning for science.\n
LOCATION:East 2/ West Hub
END:VEVENT
END:VCALENDAR