BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Fine-tuning foundation models of materials interatomic potentials 
 with frozen transfer learning  - Mariia Radova\, University of Warwick\, U
 K
DTSTART:20251201T143000Z
DTEND:20251201T150000Z
UID:TALK239764@talks.cam.ac.uk
CONTACT:Dr Fabian Berger
DESCRIPTION:Machine-learned interatomic potentials are revolutionising ato
 mistic materials simulations by providing accurate and scalable prediction
 s within the scope covered by the training data. However\, generation of a
 n accurate and robust training data set remains a challenge\, often requir
 ing thousands of first-principles calculations to achieve high accuracy. F
 oundation models have started to emerge with the ambition to create univer
 sally applicable potentials across a wide range of materials. While founda
 tion models can be robust and transferable\, they do not yet achieve the a
 ccuracy required to predict reaction barriers\, phase transitions\, and ma
 terial stability. This work demonstrates that foundation model potentials 
 can reach chemical accuracy when fine-tuned using transfer learning with p
 artially frozen weights and biases. For two challenging datasets on reacti
 ve chemistry at surfaces and stability and elastic properties of tertiary 
 alloys\, we show that frozen transfer learning with 10-20% of the data (hu
 ndreds of datapoints) achieves similar accuracies to models trained from s
 cratch (on thousands of datapoints). Moreover\, we show that an equally ac
 curate\, but significantly more efficient surrogate model can be built usi
 ng the transfer learned potential as the ground truth. In combination\, we
  present a simulation workflow for machine learning potentials that improv
 es data efficiency and computational efficiency.\n(https://www.nature.com/
 articles/s41524-025-01727-x)\n
LOCATION:https://zoom.us/j/92447982065?pwd=RkhaYkM5VTZPZ3pYSHptUXlRSkppQT0
 9
END:VEVENT
END:VCALENDAR