 |
TITLE: On Pricing Data Bundles based on Statistical Value
SPEAKER: Mokshay Madiman (Yale University)
DATE: 2:00 - 3:00 PM, Thursday, August 27, 2009
LOCATION: Pavilion I, 4U
ABSTRACT:
On pricing data bundles based on statistical value Abstract: Suppose
there are N vendors, each of whom have collected a particular data
set, and that potential customers can choose what data bundle (subset
of the N data sets) they want to buy. Our goal is to study such
situations when the value of any data bundle (or equivalently, the
amount that a customer is willing to pay for that bundle) arises from
its use in addressing some statistical question, like hypothesis
testing or estimation of a common underlying parameter. For instance,
in the case of hypothesis testing, one may use the optimal error
exponent obtainable using a particular data bundle (namely, the
relative entropy) as a measure of the statistical value of that
bundle. After formulating several versions of this problem rigorously,
we consider the following questions:
(1) Given the values associated with arbitrary data bundles, when
would the self-interested vendors (individually and in subgroups)
decide that it is in their advantage to cooperate and sell all the
data as one grand bundle?
(2) If the vendors have decided to all cooperate, how would they
allocate the revenue from selling the data cooperatively among
themselves in a way that none of them feel cheated?
For the first question, we show in three interesting model settings
that all vendors indeed would decide to form a grand coalition. For
the second question, we give explicit solutions in one setting, and
discuss possible solutions for the other two. Our analysis draws tools
from cooperative game theory, information-theoretic inequalities, and
statistical decision theory.
BIOGRAPHY:
Mokshay M. Madiman was born and raised in India. He received the
B.Tech. degree in electrical engineering from the Indian Institute of
Technology, Bombay, in 1999, and the Sc.M. and Ph.D. degrees in
applied mathematics from Brown University, Providence, RI, in 2001 and
2005 respectively. He joined the Department of Statistics at Yale
University, New Haven, CT, in July 2005, as a Gibbs Assistant
Professor. Since July 2006, he has been an Assistant Professor in the
same department, and also has a courtesy appointment in the Yale
Applied Mathematics Program. For the period of January to May 2007, he
was a Visiting Fellow at the Tata Institute of Fundamental Research,
Mumbai, India. Professor Madiman's research interests span information
theory, statistical inference, and aspects of discrete mathematics and
theoretical and applied probability.
|
|
|
printable version
