From: Alan F. Karr
[[email protected]]
Sent: Wednesday, March 15, 2006 2:31 PM
To:
[email protected]
Subject: Upcoming
Events
NDHS-DCers,
First, thanks to all, especially Larry, Sesa, Jay,
Anna, Francisco, Lisa and Mi-Ja, who helped make Monday's mid-program workshop a
success. I thought it was a really great day.
Second, I hope no one
showed up, either in person or by phone, for yesterday's cancelled working group
meeting.
Third, we will meet as scheduled next Tuesday, 3/21, at 9:15 AM.
Jim, Bahjat, Larry and I have been discussing (to be fair, in case they
want to disavow involvement, I have been discussing and they have been
listening) an idea for a working group project (probably consuming the remainder
of the semester) that I will present on Tuesday. Briefly, the idea is to try to
write down the abstractions and definitions (of data, legitimate user, intruder,
risk, utility, cost, false positive, external knowledge, ...) in a framework
that would catalyze research with true impact on the kinds of issues in Diane
Lambert's paper, also taking into account the modern electronic world. The
product of this effort would be a review/ideas paper that would fit perfectly in
the new (and not-yet-publically-announced) Journal of Privacy and
Confidentiality (of which Steve Fienberg, Cynthia Dwork of Microsoft Research
and I are the founders).
Fourth, Larry Cox will be visiting NISS/SAMSI on
Friday, 3/31, and will give a talk that is a deeper version mathematically of
his talk on 3/13 (cute date reversal, huh?). The talk will be at 11 AM in Room
104 at NISS. The abstract is below. Larry will also be available from informal
discussions that day. I hope most of the working group will be able to attend;
we can decide next Tuesday whether Larry's talk will replace the 3/28 working
group meeting.
--- Alan
MATHEMATICAL MODELS FOR DATA
SECURITY PROBLEMS IN TABLES
Lawrence
H. Cox
National
Center for Health Statistics
[email protected]
Tables are a staple of statistical analysis.
Multi-dimensional tables are becoming of increasing interest, e.g.,
high-dimensional contingency tables forming the basis of a statistical data base
query system. Data security problems in tables may be modeled
in the language of mathematical programming. (1) Rounding
some or all table entries to a fixed integer rounding base.
(2) Data suppression.
(3) Controlled adjustment of table entries, combined with
quality-preserving techniques. (4) Verifying confidentiality
protection in tables by means of a confidentiality audit.
Results based on mathematical programming theory demonstrate
that in general these models are expected to be extremely difficult to compute
(NP-hard) and in some cases may fail to have a solution. The principal exception
is a class of tables corresponding to mathematical networks, a class that
includes two-way tables. In this talk, we present
mathematical programming models for confidentiality problems in tables, describe
the class of tables of network-type, and illustrate difficulties in modeling and
computing confidentiality protection or verification outside this class.
--
******************************************************************
* Alan F. Karr, Director * Tel: 919.685.9300 *
* National Institute of Statistical Sciences * FAX: 919.685.9310 *
* 19 T. W. Alexander Drive (FedEx/UPS) * [email protected] *
* P.O. Box 14006 (USPS) * www.niss.org *
* Research Triangle Park, NC 27709-4006 * *
******************************************************************