BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Polymathic AI: Foundation Models for Science - Miles Cranmer - DAM
 TP/IoA
DTSTART:20231025T130000Z
DTEND:20231025T140000Z
UID:TALK207208@talks.cam.ac.uk
CONTACT:James Fergusson
DESCRIPTION:n the last few years\, natural language processing and compute
 r vision have experienced a fundamental shift in the way these fields use 
 machine learning. Rather than training neural networks from a randomly ini
 tialized set of parameters\, researchers have often found superior perform
 ance can be achieved by fine-tuning a general pre-trained “foundation mo
 del” trained on vast amounts of diverse data – perhaps because this mo
 del comes with better “priors” than an untrained network. Polymathic A
 I1 is a new research collaboration that aims to usher in the same shift in
  machine learning for scientific datasets. In this talk I will present the
  motivations behind the collaboration and describe the findings of our thr
 ee new papers in this space\, which examine: better numerical encodings fo
 r large language models2\, contrastive embeddings for multi-modal scientif
 ic data3\, and building machine learning models that learn from multiple t
 ypes of physics4.\n \n1 https://polymathic-ai.org/\n2https://arxiv.org/abs
 /2310.02989\n3https://arxiv.org/abs/2310.03024\n4https://arxiv.org/abs/231
 0.02994
LOCATION:Maxwell Centre
END:VEVENT
END:VCALENDAR