[Ml-stat-talks] Wilks Statistics Seminar on Friday: Sasha Rakhlin (University of Pennsylvania)

Lucy Xia lxia at princeton.edu
Thu Oct 3 15:27:47 EDT 2013

For tomorrow's wilks seminar, we will have prof. Sasha Rakhlin from Upenn,
will be a nice talk. Please find detailed information below.

=== Wilks Statistics Seminar ===

DATE:   Friday, Oct. 4

TIME:   12:30pm

LOCATION:   Sherrerd Hall 101

SPEAKER:   Sasha Rakhlin, University of Pennsylvania

TITLE:   Learning and estimation: separated at birth, reunited at last

ABSTRACT: We consider two scenarios: (a) random design regression with
square loss under the assumption that the model F is well-specified,
and (b) distribution-free statistical learning with respect to a
reference class F. The former problem is often studied in the
literature on nonparametric estimation, while the latter falls within
the purview of statistical learning theory. It is recognized in both
communities that complexity of the class F plays the key role in
determining the minimax rates: the importance of entropy in the study
of estimation goes back to Le Cam, Ibragimov and Khas'minskii, and
Birge; within the setting of statistical learning, the importance of
entropy was established in the work of Vapnik and Chervonenkis and in
subsequent works on uniform LLN within empirical process theory. But
do minimax rates for these two problems really differ? The question,
which boils down to understanding well-specified and misspecified
models, will be addressed in this talk.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/ml-stat-talks/attachments/20131003/b4696371/attachment.html>

More information about the Ml-stat-talks mailing list