|
|
|
|
|
Topics in Privacy |
The following schedule and descriptions are tentative. Topics are usually not posted earlier than the week before.
[Summer 2003] [Summer 2004] [Data Privacy Lab Projects] [Data Privacy Lab]
Abstracts of Talks and Discussions
In this talk I present a partial solution to this problem---CUES,
Checking User Expectations about Semantics. CUES is a method and a
prototype implementation for making user expectations precise and for
checking these precise expectations. CUES treats the precise
expectations as a proxy for missing specifications. It checks the
precise expectations to detect semantic anomalies---data feed behavior
that does not adhere to these expectations.
Three case studies and a validation study, all with real-world data,
provide evidence of the practicality and usefulness of CUES. The
studies indicate that a user of CUES gets substantial benefit for a
modest investment of time and effort. In addition to automated detection
of anomalies, the benefit often includes a better understanding of the
user's own expectations, of the data feeds, and of existing and missing
documentation.
The rest of the talk will be devoted to brainstorming and
discussion of some potentially new areas in the intersection
of HCI, systems, privacy, and security.
Dr. Shamos' widely cited paper on voting system examination, written in 1993, is available at: http://www.cpsr.org/conferences/cfp93/shamos.html
Information about Dr. Shamos' E-voting class is available at:
http://euro.ecom.cmu.edu/program/courses/tcr17-803/
The definition of privacy and the requirements for a safe sanitization
are important contributions of this work. Our definition of privacy
formalizes the notion of protection from being brought to the
attention of others -- one's privacy is maintained to the extent that
one blends in with the crowd. Our definition of a safe sanitization
emulates the definition of semantic security for a cryptosystem and
says, roughly, that an adversary given access to the sanitized data is
not much more able to compromise privacy than an adversary who is not
given access to the sanitization.
Joint work with Shuchi Chawla, Frank McSherry, Adam Smith, and Hoeteck Wee.
Much of the software people use for everyday purposes incorporates
elements developed and maintained by someone other than their
integrator. These elements include not only code but also data feeds.
Although everyday information systems are not mission critical, they
must be dependable enough for practical use. This is limited by the
dependability of the incorporated elements.
It is particularly difficult to evaluate the dependability of data
feeds. The specifications of data feeds are often even sketchier than
the specifications of software components, the data feeds may be changed
by their proprietors, and everyday users of data feeds only have enough
knowledge about the application domain to support their own usage. These
factors inhibit many dependability enhancement techniques, which require
a model of proper behavior for failure detection, preferably in the form
of specifications.
[Presenter: Orna Raz]
In the first part of this talk, Dr. Hong will go over some of his past
work in understanding and developing privacy-sensitive ubicomp
systems, primarily focusing on location privacy. He will also draw
on interesting themes and connect them with his current
research here at CMU in developing and deploying privacy-sensitive
ubicomp systems.
[Presenter: Jason Hong]
Dichotomies between privacy attitudes and actual behavior have been noted in the literature but not yet fully explained. Individuals often claim to be highly concerned about their personal privacy, but few adopt technologies to protect it, and many release personal information in exchange for small rewards. We apply lessons from the research on behavioral economics to understand the individual decision making process with respect to privacy. We show that it is unrealistic to expect individual rationality in this context. Models of self-control problems and immediate gratification offer more realistic descriptions of the decision process and are more consistent with currently available data. In particular, we show why individuals who may genuinely want to protect their privacy might not do so because of psychological distortions well documented in the behavioral literature; we show that these distortions may affect not only ‘naive’ individuals but also ‘sophisticated’ ones; and we show that this may occur also when individuals perceive the risks from not protecting their privacy as significant. Finally, we present results from a survey of individual privacy attitudes and behavior that we conducted in May 2004 to test our theoretical model.
[Presenter: Alessandro Acquisti]
It is two weeks before the U.S. presidential election; what is the state of electronic voting machines? Will there be another Florida disaster or will the U.S. be lucky if just one state has a problem? In this talk, Michael Shamos will bring us up-to-date on the state of affairs regarding legal actions and challenges to electronic voting machines and how these problems are likely to impact the upcoming election.
[Presenter: Michael Shamos]
We will describe definitions and algorithmic results for the ``census
problem''. Informally, in a census individual respondents give
private information to a trusted (and trustworthy) party, who
publishes a sanitized version of the data. There are two
fundamentally conflicting requirements: privacy for the respondents
and utility of the sanitized data. Unlike in the study of secure
function evaluation, in which privacy is preserved to the extent
possible given a specific functionality goal, in the census problem
privacy is paramount; intuitively, things that cannot be learned
"safely" should not be learned at all.
[Presenter: Cynthia Dwork, Microsoft Research]
At least once each semester, we devote a TIP session to brainstorming and reflecting on the hottest issues in privacy areas and the many new and provocative ideas posed by speakers in earlier TIPs during the semester. In this session, open and candid discussion is particularly welcomed. Among the topics are: privacy-preserving data mining, privacy in electronic voting, economics of privacy, privacy in ubiquitous computing, formal protection models, privacy-preserving surveillance, and/or any other topic of interest to those in attendance. All are welcomed to come, listen, and participate.
We summarize three recent results in anonymous communication
technologies.
The talk will assume far less prior knowledge in anonymous communication
than this abstract does.
[Presenter: Mike Reiter]
We begin by describing some of the principles underlying the "statistical approach"
to confidentiality and disclosure limitation, especially the risk-utility tradeoff,
and we explain by example our statistical notion of utility. We then contrast this
with "privacy preserving" methods now popular in the data mining literature.
This will serve as a prelude to a brainstorming discussion about the viability
of the notion of "selective revelation" that has been proposed as a fundamental
component of the DARPA-led Total Information Awareness Program, or
Terrorism Information Awareness program as it is now known.
[Presenter: Stephen E. Fienberg]
For the past two years, the Markle Foundation has undertaken a series
of studies on National Security through it's panel on National
Security. The results have had significant impact on the structure of
DHS and the current Intelligence debate. Three key recent documents and
the url to the main web page:
The proposed direction raised a great deal of privacy issues which we
will brainstorm in this meeting.
[Presenter: David Farber]
Our protocols for two parties are comparable to those of Freedman,
Nissim, and Pinkas~, except in the malicious case, in which we achieve
security without the use of the expensive cut-and-choose technique.
Freedman, Nissim, and Pinkas also address the problem of private
set-intersection for multiple parties, giving a protocol secure
against honest-but-curious parties with O(n^2k) total communication
complexity, but no protocol secure against malicious adversaries. We
give a protocol for this scenario secure against honest-but-curious
parties with O(cnk) total
communication complexity, and against malicious parties with
communication complexity O(n^2k^2).
There was no previous efficient protocol for the problems of
cardinality set-intersection (for n >= 3 players or n >= 2
malicious players), threshold set-intersection, or over-threshold
set-intersection. We give the first efficient protocols for these
problems without generic secure circuit computation.
Note: This is a mathematically-oriented talk. Background math and
cryptography will be covered at the beginning of the talk.
A draft of the paper is available
here.
Many factors have contributed to this evolution of Web information; a crucial one is the way in which technological and social networks have become intertwined, enabling new topics, ideas, and behaviors to cascade along an underlying social network of interaction that parallels the hyperlinks of the Web itself. In this talk, I will discuss some of the current approaches to the challenges that are raised by these issues, including new algorithms to analyze the rich array of temporal phenomena that abound in current information networks, and search techniques that can handle the temporal fluctuations and bursty dynamics of on-line topics.
In this talk we consider the problem of privately computing the
set-intersection (private matching) of sets, as well as several
variations on this problem: cardinality set-intersection, threshold
set-intersection, and over-threshold set-intersection. Cardinality
set-intersection is the problem of determining the size of the
intersection set, without revealing the actual set. In threshold
set-intersection, only the elements which appear at least a threshold
number t times in the players' private inputs are revealed.
Over-threshold set-intersection is a variation on threshold
set-intersection in which not only the threshold set is revealed, but
also the popularity of each element in the set. Let there be n >= 2
players, c < n dishonestly colluding, each with a private input set of
k elements.
[Presenter: Lea Kissner]
The working metaphor for the Web in its early days was that of a `universal encyclopedia,' a repository containing vast amounts of human knowledge. More recently, our view has shifted to incorporate more dynamic forms of information and a more explicit `time axis' -- the Web as a current-awareness medium, supporting on-line news, chat, discussion, commentary, and emerging media such as weblogs.
[Presenter: Jon Kleinberg]
Related Data Privacy links