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:Machine learning in Physics\, Chemistry and Materi
als discussion group (MLDG)
SUMMARY:Neural Network Approximations for Calabi-Yau Metri
cs - Challenger Mishra
DTSTART;TZID=Europe/London:20210222T170000
DTEND;TZID=Europe/London:20210222T173000
UID:TALK157639AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/157639
DESCRIPTION:String theory is the only known consistent theory
of quantum gravity. The extra-dimensional part of
space posited by string theory is often described
by complex geometries called Calabi--Yau manifolds
. In order for string theory to make predictions f
or masses of fundamental particles\, such as elect
rons\, we require knowledge of a special Riemannia
n metric over Calabi--Yau threefolds. Such metrics
\, known as Ricci flat metrics\, are solutions to
partial differential equations that are notoriousl
y difficult to solve. In fact\, no analytic soluti
on is known for metrics of Calabi--Yau threefolds.
\nWe employ techniques from machine learning to de
duce numerical flat metrics for certain phenomenol
ogically important Calabi--Yau geometries\, namely
\, the Fermat quintic\, the Dwork quintic\, and th
e Tian-Yau manifold. We show that measures that as
sess the Ricci flatness of the geometry decrease a
fter training by three orders of magnitude. This i
s corroborated on the validation set\, where the i
mprovement is more modest. Finally\, we demonstrat
e that discrete symmetries of manifolds can be lea
rned in the process of learning the metric.
LOCATION:virtual ZOOM meeting ID: 263 591 6003\, Passcode:
000042\, https://us02web.zoom.us/j/2635916003?pwd=
ZlBEQnRENGwxNmJGMENGMWxjak5nUT09
CONTACT:Bingqing Cheng
END:VEVENT
END:VCALENDAR