[bull-ia] PGMO lecture series of Tim Roughgarden, Data-Driven Optimal Auction

Content-Type: text/html;
charset=utf-8

Dear Colleagues,

The Gaspard Monge Programme for Optimisation =
and Operations Research
(PGMO) is pleased to advertise the =
exceptional lecture series of

Professor Tim =
ROUGHGARDEN, from Stanford University

on

        DATA-DRIVEN =
OPTIMAL AUCTION THEORY

at Ecole =
polytechnique, Palaiseau, September, 14-21, 2017.

This is jointly organized by PGMO, Ecole polytechnique, and =
ENS Paris
Saclay.

This lecture is intended to graduate students and to =
researchers
(academics or from industry).

The schedule is the following

Lecture 1 – Thursday, Sept 14, 10h30-12h30, Amphi Carnot
Lecture 2 – Friday, Sept 15, 10h30-12h30, Amphi Carnot
Lecture 3 – Tuesday, Sept 19, 10h30-12h30, Amphi Carnot
Lecture 4 – Wednesday, Sept 20, 10h30-12h30, Amphi =
Lagarrigue
Lecture 5 – Thursday, Sept 21, 10h30-12h30, =
Amphi Carnot

Registration (free of charge) =
is requested; register by filling
the form:

https://www.fondation-hadamard.fr/fr/content/pgmo-course-regist=
ration-tim-roughgarden

Owing to « vigiepirate » security measures, attendees may =
have to
show an identification document.

To access to school:

* =
Directions to Ecole Polytechnique, Palaiseau (Saclay Campus)
https://www.polytechnique.edu/en/maps-and-directions?language=3D=
en

* Attendees should go first to the =
« Accueil » desk of the school,
to get access to the =
lecture rooms area.

See the map of =
amphis:
http://softs.polytechnique.fr/cpm/avant_projets/plans/plan_pc_a=
mphis.html

For =
information, contact magali.lechaponnier@fondation-hadamard.fr
Scientific organizers:

P. =
Carpentier (ENSTA)
S. Charousset (EDF)
S. =
Gaubert (INRIA and Ecole polytechnique)
V. Perchet (ENS =
Paris Saclay)
F. Santambrogio (Universit=C3=A9 Paris =
Sud)
T. Tomala (HEC)

https://www.fondation-hadamard.fr/PGMO
—————————————————————=
—

Summary of the Lecture

The traditional economic approach to =
revenue-maximizing auction design
posits a known prior =
distribution over what bidders are willing to
pay, and =
then solves for the auction that maximizes the seller’s
expected revenue with respect to this distribution. But where =
does
this distribution come from? One obvious answer is =
from data, in the
form of previous bids by comparable =
bidders in auctions for comparable
items.  The goal =
of this short course is to develop theory to help
reason =
about questions such as: (i) for much data is necessary and
sufficient to identify a near-optimal auction? (ii) what is =
the
optimal way to use data? (iii) are there fundamental =
trade-offs
between auction complexity and auction =
optimality?

The course will =
include introductions to auction theory and learning
theory =
(with no prior knowledge assumed, just mathematical maturity),
before proceeding to cutting-edge (mostly 2016–2017) =
research at
their intersection.

Detailed schedule (five =
two-hour lectures):

Lecture =
1 (Thursday, September 14): Introduction to auction
theory.  Second-price auctions and dominant =
strategies.  Myerson’s
theory of revenue-maximizing =
auctions.  The Bulow-Klemperer theorem and
approximation guarantees for simple auctions.

Lecture 2 (Friday, September 15): =
Introduction to learning
theory.  Statistical =
learning theory and the PAC model.  Measuring
hypothesis complexity (pseudodimension, etc.).  =
Radamacher complexity.
Examples.

Lecture 3 (Tuesday, September 19): Learning =
near-optimal
auctions in the batch model.  The =
pseudodimension of classes of simple
auctions.  =
Compression schemes.  Computational considerations.

Lecture 4 (Wednesday, September 20): Learning =
near-optimal
auctions online.  Online models of =
learning.  The
« follow-the-regularized-leader » =
algorithm.  Online-to-offline reductions.

Lecture 5 (Thursday, September =
21): The cutting edge:
Multi-item auctions; new =
concentration bounds for product distributions;
strategic =
issues in data collection.

=