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:Falkon: fast and optimal kernel method for large s
cale machine learning - Alessandro Rudi (INRIA\; E
NS - Paris)
DTSTART;TZID=Europe/London:20180119T090000
DTEND;TZID=Europe/London:20180119T094500
UID:TALK97900AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/97900
DESCRIPTION:Kernel methods provide a principled way to perform
non linear\, nonparametric learning. They rely on
solid functional analytic foundations and enjoy o
ptimal statistical properties. However\, at least
in their basic form\, they have limited applicabil
ity in large scale scenarios because of stringent
computational requirements in terms of time and es
pecially memory. In this paper\, we take a substan
tial step in scaling up kernel methods\, proposing
FALKON\, a novel algorithm that allows to efficie
ntly process millions of points. FALKON is derived
combining several algorithmic principles\, namely
stochastic subsampling\, iterative solvers and pr
econditioning. Our theoretical analysis shows that
optimal statistical accuracy is achieved requirin
g essentially O(n) memory and O(n sqrt{n}) time. A
n extensive experimental analysis on large scale d
atasets shows that\, even with a single machine\,
FALKON outperforms previous state of the art solut
ions\, which exploit parallel/distributed architec
tures.
LOCATION:Seminar Room 1\, Newton Institute
CONTACT:INI IT
END:VEVENT
END:VCALENDAR