Abstracts for Privacy Day at the Weizmann Institute
Wednesday, July 12th 2006 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

SPEAKER: Yehuda Lindell, Bar-Ilan University 
TITLE: Secure Multiparty Computation and Privacy

ABSTRACT:

In this talk, we will present a short tutorial on the basic concept of
secure multiparty computation, with a focus on definitions and modeling. In
addition to presenting the high-level ideas behind the definition, we will
motivate why security is defined in this way and what it implies. We will
also talk about how secure computation can be used to improve privacy, and
what the limitations of the approach are.


SPEAKER: Adam D. Smith, Weizmann Institute of Science
TITLE: Pinning Down "Privacy"

ABSTRACT:

This talk looks at some of the approaches taken to defining privacy
and confidentiality in statistical databases. Consider a trusted
agency that collects individuals' data and would like to make "global"
statistics about the population available to the public, while
keeping "individual" information secret. I'll discuss some of the ways
that people have formalized "global" and "individual" in this context,
their relations to each other, and some of the problems they come
with. The focus will be on mathematically rigorous definitions. My
background will probably bias the discussion towards ideas that have
come from the computer science literature.


SPEAKER: Yosi Rinott, Hebrew University
TITLE: Estimating disclosure risk measures using local smoothing.

ABSTRACT:

A file to be released by an agency consists of a sample from a population in the
form of a frequency table,  and the agency wants to 
assess potential disclosure risk.
Focusing only on key variables in the table, we consider risk measures
which are based on small population cells which are represented in the sample.

Using the structure of the sample table, and statistical assumptions on how it
was generated, we estimate the population entries associated with small sample cells.
We present a local smoothing method, estimate its parameters using    rather
general statistical model, and compare this approach to existing literature.

Joint work with Natalie Shlomo


SPEAKER: Kobbi Nissim, Ben-Gurion
TITLE: Calibrating Noise to Sensitivity in Private Data Analysis

ABSTRACT:

In this talk we discuss three types of results related to privacy in
statistical databases. First, we briefly present a
cryptographically flavored definition of privacy in statistical databases
that captures the intuition that whatever information is released should not depend strongly
on any single individual's data.

In the second part of the talk, we present a framework for output
perturbation that formalizes the intuition that it is safe to release answers to
queries that are not too sensitive to small 'local' changes in the database.
We define a combinatorial measure of sensitivity and show that by properly
calibrating the perturbation noise to sensitivity we get provably private
sanitization mechanisms. We demonstrate the usefulness of this approach for
several types of queries.

Finally, we will present separation results between basic models of
sanitization. In particular, we will show the increased value of interactive
sanitization mechanisms over non-interactive mechanisms.


SPEAKER:  Shmuel Onn, Technion - Israel Institute of Technology
TITLE: Multiway Tables: Universality and Optimization

ABSTRACT:

I will describe our recent work on universality of short 3-way
tables and convex integer programming over long d-way tables,
and discuss possible implications on secure margin disclosure.
For instance, an immediate consequence of our work is that
detecting uniqueness of an entry value in all tables with same
margins as those disclosed from a source table in a data base
is hard for short 3-way tables and easy for long d-way tables.


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%