Signal Processing - January 2017 - 94

lecture notes
S.Y. Kung

Compressive Privacy: From Information/Estimation
Theory to Machine Learning

M

ost of our daily activities are now
moving online in the big data era,
with more than 25 billion devices
already connected to the Internet, to possibly over a trillion in a decade. However, big data also bears a connotation of
"big brother" when personal information
(such as sales transactions) is being ubiquitously collected, stored, and circulated
around the Internet, often without the
data owner's knowledge. Consequently, a
new paradigm known as online privacy
or Internet privacy is becoming a major
concern regarding the privacy of personal and sensitive data.
As depicted in Figure 1, Internet
data live in two different worlds: 1) the
private sphere, where data owners generate and process decrypted data, and
2) the public sphere, where the data in
cloud servers are presumably encrypted
and therefore unaccessible by intruders.
However, the data may be decrypted by
the intended and trusted "authorities,"
who will be provided with the right key
for the data decryption. Following the
cryptographic channel and adversarial
models, formally defined by Claude
Shannon, we shall also name the data
owner, intended user, and intruder as
Alice, Bob, and Eve, respectively. Recall
that the typical security protocol hinges
upon Alice's passing a decrypting key
to Bob but not to Eve. Since the notion
of an unbreakable key is questionable,
there is no wonder that Internet data

Digital Object Identifier 10.1109/MSP.2016.2616720
Date of publication: 11 January 2017

94

remain highly vulnerable to unauthorized leakages and hacker attacks.

Data owner should have
control over data privacy
Privacy-preserving (PP) tools have a
broad spectrum of applications, covering numerous types of Internet data
(such as speech, image, location, and
media/social/health data). They all
require a delicate balance between utilization and privacy. For example, in

case of a bomb explosion, images from
various mobile sources near the crime
scene may be collected by authorities for
wide-scale forensic analysis. Ideally, the
uploaded images should provide critical
and relevant information to help capture
the targeted suspects while protecting
the full facial images of the innocent
from being leaked to the public.
Data are not just a collection of
words/numbers working in isolation,
rather they encompass the global and

Public Space: Cloud
Cloud
Server
Intruder

Trusted Authority
Encrypted
Data

Adversary

Decrypted
Data

Private Space: Clients

Figure 1. In collaborative learning environments, individual data are uploaded to the cloud. From the
privacy perspective, data in the private sphere versus the public sphere should be treated differentially,
which calls for a novel PP encoding paradigm, known as compressive privacy (CP). For privacy protection, the query data uploaded to the public sphere should be designed to retain the information
useful for the intended application and should not be easily repurposed into malicious exploitation.
IEEE Signal Processing Magazine

|

January 2017

|

1053-5888/17©2017IEEE
Table of Contents for the Digital Edition of Signal Processing - January 2017
