BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Letting the machines vote: Committee neural network potentials con
 trol generalization errors and enable active learning - Christoph Schran (
 UCL)
DTSTART:20201104T113000Z
DTEND:20201104T123000Z
UID:TALK153211@talks.cam.ac.uk
CONTACT:Angela Harper
DESCRIPTION:Machine learning has emerged in recent years as a powerful\nto
 ol for the description of complex chemical systems.\nBased on the well-est
 ablished Behler-Parrinello neural\nnetwork potential methodology for inter
 atomic potentials\,\nI show in this talk that it has multiple benefits to\
 ncombine a number of machine learning models to a committee.\nInstead of a
  single model\, multiple models yield an average\nthat outperforms its ind
 ividual members as well as a measure\nof the generalization error in the f
 orm of the committee\ndisagreement.\n\nThis  disagreement can not only be 
 used  to  identify  the\nmost relevant  configurations  to  build  up  the
   model’s\ntraining set in an active learning procedure\, but can also\n
 be monitored and biased during simulations to control the\ngeneralization 
 error.\n\nThis facilitates the adaptive development of committee\nneural n
 etwork potentials and their training sets\, while\nkeeping the number of a
 b initio calculations to a minimum.\nIn this talk I will show that this me
 thodology enables the\nrapid development of robust and accurate machine le
 arning\npotentials for complex aqueous systems.\n\nJoin Zoom Meeting:\n* h
 ttps://bham-ac-uk.zoom.us/j/85702415099?pwd=VTh5aFZ4Sm9Nc1dxZXYwelJpc1JtZz
 09\n* Meeting ID: 857 0241 5099\n* Passcode: 501623
LOCATION:Zoom
END:VEVENT
END:VCALENDAR