BEGIN:VCALENDAR
VERSION:2.0
PRODID:talks.ox.ac.uk
BEGIN:VEVENT
SUMMARY:Gaussian Transformations\, Recursive Learning\, and the Maximum Li
kelihood Estimation of Conditional Distribution Functions (joint with Sami
Stouli\, Bristol) - Richard Spady (Nuffield College)
DTSTART;VALUE=DATE-TIME:20191025T141500
DTEND;VALUE=DATE-TIME:20191025T153000
UID:https://talks.ox.ac.uk/talks/id/11ff8bbe-831f-43bd-bca5-ec41ceb4f155/
DESCRIPTION:We propose an algorithm for machine learning of conditional di
stribution functions for a dependent variable (Y ) with continuous support
. The algorithm produces a complete description of the conditional distrib
ution function at all observed points in the covariate (X) space\, and pro
vides a similar estimate for other possible covariate values. The descript
ions it provides are quite general and are globally valid conditional dens
ities.The algorithm is multilayered and feed-forward. Each layer has the s
ame statistical interpretation: Layer k takes a vector e(k-1) that is near
ly perfectly marginally Gaussian and makes it more marginally Gaussian and
more independent of X. It does this by applying a continuous monotonic tr
ansformation that varies depending on an observation’s X value. Each lay
er is estimated by an elastic net regularization of maximum likelihood. We
demonstrate Wilks’ phenomenon for the composite algorithm and show how
to calculate the algorithm’s effective dimension.\n\nPlease sign up for
meetings here: https://docs.google.com/spreadsheets/d/1qPQqXivNYBDNJY_0OdH
cZfjslHLu5UtSVQMd0LpETqc/edit#gid=0\nSpeakers:\nRichard Spady (Nuffield Co
llege)
LOCATION:Manor Road Building (Seminar Room C)\, Manor Road OX1 3UQ
URL:https://talks.ox.ac.uk/talks/id/11ff8bbe-831f-43bd-bca5-ec41ceb4f155/
BEGIN:VALARM
ACTION:display
DESCRIPTION:Talk:Gaussian Transformations\, Recursive Learning\, and the M
aximum Likelihood Estimation of Conditional Distribution Functions (joint
with Sami Stouli\, Bristol) - Richard Spady (Nuffield College)
TRIGGER:-PT1H
END:VALARM
END:VEVENT
END:VCALENDAR