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:Computer Laboratory Systems Research Group Seminar
SUMMARY:DataFlow SuperComputing for BigData - Veljko Milut
inovic
DTSTART;TZID=Europe/London:20170928T150000
DTEND;TZID=Europe/London:20170928T160000
UID:TALK72573AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/72573
DESCRIPTION:This presentation analyses the essence of DataFlow
SuperComputing\, defines\nits advantages and shed
s light on the related programming model. DataFlow
\ncomputers\, compared to ControlFlow computers\,
offer speedups of 20 to 200\n(even 2000 for some a
pplications)\, power reductions of about 20\, and
size\nreductions of also about 20. However\, the p
rogramming paradigm is\ndifferent\, and has to be
mastered. The talk explains the paradigm\, using\n
Maxeler as an example\, and sheds light on the ong
oing research in the\nfield. Examples include: Cry
ptography and Security\, Trading and Finances\,\nC
reditDerivatives and numerous related BankingAppli
cations\,\nSignalProcessing\, GeoPhysics\, Weather
Forecast\, OilGas\, DataEngineering\,\nDataMining\
, SmartGrid\, ScientificSimulations\, BrainResearc
h\, Genomics\,\netc.\n\nAlso\, a recent study from
Tsinghua University in China is presented\, which
\nreveals that\, for Shallow Water Weather Forecas
t (a BigData problem)\, on\nthe 1U level\, the Max
eler DataFlow machine is 14 times faster than the\
nTianhe machine\, rated #1 on the Top 500 list (ba
sed on Linpack\, which is a\nsmalldata benchmark).
Given enough time\, the talk also gives a tutoria
l\nabout the programming in space\, which is the p
rogramming paradigm used for\nthe Maxeler dataflow
machines (established in 2014 by Stanford\, Imper
ial\,\nTsinghua\, and the University of Tokyo). Th
e talk concludes with selected\nexamples and a too
l overview (appgallery.maxeler.com and\nwebIDE.maxeler.com). A det
ailed tutorial on programming in space will be\nav
ailable after the talk. Related hands-on activitie
s will be performed by\nremote login (maxeler.mi.sanu.ac.rs
a>). Since December 2016\, Maxeler is also\navaila
ble via Amazon AWS. In December 2016\, Hitachi of
Japan announced its\npartnership with Maxeler (als
o available via Amazon AWS)\, stating that\,\nfor
their finance and cryptography applications\, Maxe
ler is orders of\nmagnitude faster than any other
ControlFlow platform (i.e.\, CPU or GPU).\n\n\nBio
: Prof. Veljko Milutinovic (1951) received his PhD
from the University of\nBelgrade\, spent about a
decade on various faculty positions in the USA\n(m
ostly at Purdue University)\, and was a co-designe
r of the DARPAs first\nGaAs RISC microprocessor. L
ater\, for almost 3 decades\, he taught and\ncondu
cted research at the University of Belgrade\, in E
E\, BA\, MATH\, and\nPHYS /CHEM. Now he serves as
the Chairman of the Board for the Maxeler\noperati
on in Belgrade\, Serbia. His research is mostly in
datamining algorithms and\ndataflow computing\, w
ith the emphasis on mapping of data analytics\nalg
orithms onto fast energy efficient architectures.
For 7 of his books\,\nforewords were written by 7
different Nobel Laureates with whom he\ncooperated
on his past industry sponsored projects. He has o
ver 40 IEEE\njournal papers\, over 40 papers in ot
her SCI journals (4 in ACM journals)\,\nover 400 T
homson-Reuters citations\, and about 4000 Google S
cholar\ncitations. Short courses on the subject he
delivered so far in a number of\nuniversities wor
ldwide: MIT\, Harvard\, Boston\, NEU\, Columbia\,
NYU\,\nPrinceton\, Temple\, Purdue\, IU\, UIUC\, M
ichigan\, EPFL\, ETH\, Karlsruhe\,\nHeidelberg\, e
tc. Also at the World Bank in Washington DC\, BNL\
, IBM TJ\nWatson\, Yahoo\, etc.\n
LOCATION:FW26\, Computer Laboratory\, William Gates Buildin
g
CONTACT:Liang Wang
END:VEVENT
END:VCALENDAR