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:Isaac Newton Institute Seminar Series
SUMMARY:Random Batch Methods for Interacting Particle Syst
ems and Consensus-based Global Non-convex Optimiz
ation in High-dimensional Machine Learning (copy)
- Shi Jin (Shanghai Jiao Tong University)
DTSTART;TZID=Europe/London:20191111T140000
DTEND;TZID=Europe/London:20191111T150000
UID:TALK134683AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/134683
DESCRIPTION:We develop random batch methods for interacting pa
rticle systems with large number of particles. The
se methods

use small but random batches for pa
rticle interactions\,

thus the computational co
st is reduced from O(N^2) per time step to O(N)\,
for a

system with N particles with binary inter
actions.

For one of the methods\, we give a par
ticle number independent error estimate under some
special interactions.

Then\, we apply these m
ethods

to some representative problems in mathe
matics\, physics\, social and data sciences\, incl
uding the Dyson Brownian

motion from random ma
trix theory\, Thomson'\;s problem\,

distribu
tion of wealth\, opinion dynamics and clustering.
Numerical results show that

the methods can cap
ture both the transient solutions and the global e
quilibrium in

these problems.

We also ap
ply this method and improve the consensus-based gl
obal optimization algorithm for high

dimension
al machine learning problems. This method does not
require taking gradient in finding global

min
ima for non-convex functions in high dimensions.**
LOCATION:Seminar Room 2\, Newton Institute
CONTACT:INI IT
END:VEVENT
END:VCALENDAR
**