BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Cambridge Ellis Unit
SUMMARY:Cambridge ELLIS seminar series – Dr Tian Xie – 23 
 Jan 2024 – 2pm - Dr Tian Xie 
DTSTART;TZID=Europe/London:20240123T140000
DTEND;TZID=Europe/London:20240123T150000
UID:TALK209986AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/209986
DESCRIPTION:The Cambridge ELLIS Unit Seminar Series holds talk
 s by leading researchers in the area of machine le
 arning and AI. Our next  speaker for 2024 will be 
 Dr. Tian Xie. Details of his talk can be found bel
 ow. \n\nTitle: “MatterGen: a generative model for 
 inorganic materials design"\n\nAbstract: The desig
 n of functional materials with desired properties 
 is essential in driving technological advances in 
 areas like energy storage\, catalysis\, and carbon
  capture. Traditionally\, materials design is achi
 eved by screening a large database of known materi
 als and filtering down candidates based on the app
 lication. Generative models provide a new paradigm
  for materials design by directly generating entir
 ely novel materials given desired property constra
 ints. In this talk\, we present MatterGen\, a gene
 rative model that generates stable\, diverse inorg
 anic materials across the periodic table and can f
 urther be fine-tuned to steer the generation towar
 ds a broad range of property constraints. To enabl
 e this\, we introduce a new diffusion-based genera
 tive process that produces crystalline structures 
 by gradually refining atom types\, coordinates\, a
 nd the periodic lattice. We further introduce adap
 ter modules to enable fine-tuning towards any give
 n property constraints with a labeled dataset. Com
 pared to prior generative models\, structures prod
 uced by MatterGen are more than twice as likely to
  be novel and stable\, and more than 15 times clos
 er to the local energy minimum. After fine-tuning\
 , MatterGen successfully generates stable\, novel 
 materials with desired chemistry\, symmetry\, as w
 ell as mechanical\, electronic and magnetic proper
 ties. Finally\, we demonstrate multi-property mate
 rials design capabilities by proposing structures 
 that have both high magnetic density and a chemica
 l composition with low supply-chain risk. We belie
 ve that the quality of generated materials and the
  breadth of MatterGen's capabilities represent a m
 ajor advancement towards creating a universal gene
 rative model for materials design.\nhttps://eng-ca
 m.zoom.us/j/81198870418?pwd=OFhHenUvM1JtWFlHRWt3aU
 12VkYxQT09 \n
LOCATION:https://eng-cam.zoom.us/j/81198870418?pwd=OFhHenUv
 M1JtWFlHRWt3aU12VkYxQT09
CONTACT:
END:VEVENT
END:VCALENDAR