Dear all,
On Friday July 7 we have Nikita Zhivotovskiy speaking in the
Statistics and Machine Learning Thematic Seminar. Nikita will be
visiting us July 3-7. He has done excellent work both in statistics
and machine learning, so it is highly recommended to come meet him.
Nikita Zhivotovskiy (UC Berkeley, Department of Statistics,
https://sites.google.com/view/nikitazhivotovskiy/)
Friday July 7, 15h00-16h00
In person, at the University of Amsterdam
Location: Science Park 904, Room A1.04
Sharper Risk Bounds for Statistical Aggregation
In this talk, we take a fresh look at the classical results in
the theory of statistical aggregation, focusing on the transition
from global complexity to a more manageable local one. The goal of
aggregation is to combine several base predictors to achieve a
prediction nearly as accurate as the best one. This flexible
approach operates without any assumptions on the structure of the
class or the nature of the target. Though aggregation is studied in
both sequential and statistical settings, each with their unique
differences, they both traditionally use the same "global"
complexity measure. Our discussion will highlight the lesser-known
PAC-Bayes localization method used in our proofs, allowing us to
prove a localized version of a classical bound for the exponential
weights estimator by Leung and Barron, and a deviation-optimal
localized bound for the Q-aggregation estimator. Furthermore, we
will explore the links between our work and ridge regression. Joint
work with Jaouad Mourtada and Tomas Vaškevičius.
Seminar organizers:
Tim van Erven
Botond Szabo
https://mschauer.github.io/StructuresSeminar/
--
Tim van Erven <tim@timvanerven.nl>
www.timvanerven.nl