[Ml-stat-talks] A tutorial on multi-armed bandits and a talk on finite stochastic optimization

Sebastien Bubeck sbubeck at Princeton.EDU
Mon Apr 30 20:50:40 EDT 2012


Hi,

this Thursday 05/03 I am giving two talks that might be of interest to the stats/machine learning people on campus:

Talk 1: Warren Powell's class ORF 569, May 3rd, 9-11am, Room 125 - Sherrerd Hall.
I will present my Tutorial on Bandits Games. See http://www.princeton.edu/~sbubeck/tutorial.html for the abstract/slides.

Talk 2: Electrical Engineering seminar series, May 3rd, 3:30-4:30pm, room B-205, Host: Sergio Verdu
Title: Two basic problems in finite stochastic optimization

Abstract: I will present a new theoretical perspective on two basic problems
arising in stochastic optimization. The first one is arguably the most
elementary problem in stochastic optimization: assume that one wants to find
the maximum of function defined on a finite set, and that one is given a
budget of n noisy evaluations. What is the best sequential allocation
procedure for the evaluations?
The second problem that I will discuss is inspired from the issue of
security analysis of a power system. We formalize this problem as follows:
Let X be a set, and A a subset of X of "interesting" elements in X. One can
access X only through requests to a finite set of probabilistic experts.
More precisely, when one makes a request to the i^th expert, the latter
draws independently at random a point from a fixed probability distribution
P_i over X. One is interested in discovering rapidly as many elements of A
as possible, by making sequential requests to the experts.

Best,
Sebastien Bubeck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/ml-stat-talks/attachments/20120501/866b65df/attachment.html>


More information about the Ml-stat-talks mailing list