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 @ CUED
SUMMARY:Unbiased Estimation of the Eigenvalues of Large Im
plicit Matrices - Professor Ryan Adams\, Princeton
DTSTART;TZID=Europe/London:20170914T110000
DTEND;TZID=Europe/London:20170914T120000
UID:TALK74501AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/74501
DESCRIPTION:\nMany important problems are characterized by the
eigenvalues of a\nlarge matrix. For example\, th
e difficulty of many optimization\nproblems\, such
as those arising from the fitting of large models
in\nstatistics and machine learning\, can be inve
stigated via the spectrum\nof the Hessian of the e
mpirical loss function. Network data can be\nunde
rstood via the eigenstructure of the Laplacian mat
rix through\nspectral graph theory. Quantum simul
ations and other many-body\nproblems are often cha
racterized via the eigenvalues of the solution\nsp
ace\, as are various dynamic systems. However\, n
aive eigenvalue\nestimation is computationally exp
ensive even when the matrix can be\nrepresented\;
in many of these situations the matrix is so large
as to\nonly be available implicitly via products
with vectors. Even worse\,\none may only have noi
sy estimates of such matrix vector products. In\n
this talk I will discuss how several different ran
domized techniques\ncan be combined into a single
procedure for unbiased estimates of the\nspectral
density of large implicit matrices in the presence
of noise.\n\n
LOCATION:CBL Room BE-438\, Department of Engineering
CONTACT:Pat Wilson
END:VEVENT
END:VCALENDAR