Understanding the Effects of Real-World Behavior in Statistical - - PowerPoint PPT Presentation

understanding the effects of real world behavior in
SMART_READER_LITE
LIVE PREVIEW

Understanding the Effects of Real-World Behavior in Statistical - - PowerPoint PPT Presentation

Understanding the Effects of Real-World Behavior in Statistical Disclosure Attacks Simon Oya , Carmela Troncoso and Fernando Prez-Gonzlez In Introduction. Mix ixes. MIX The adversary is able to learn Alices sending profile!!! 2 In


slide-1
SLIDE 1

Understanding the Effects of Real-World Behavior in Statistical Disclosure Attacks

Simon Oya, Carmela Troncoso and Fernando Pérez-González

slide-2
SLIDE 2

In

  • Introduction. Mix

ixes.

2

The adversary is able to learn Alice’s sending profile!!! MIX

slide-3
SLIDE 3

In

  • Introduction. Mix

ixes.

3

Anonymity? How?

  • Changing appearance

(re-encryption) MIX

slide-4
SLIDE 4

In

  • Introduction. Mix

ixes.

4

Anonymity? How?

  • Changing appearance

(re-encryption) MIX The adversary is able to learn Alice’s sending profile with the timing information!!!

slide-5
SLIDE 5

In

  • Introduction. Mix

ixes.

5

ANONYMOUS Anonymity? How?

  • Changing appearance

(re-encryption)

  • Removing timing

information (delays) MIX

slide-6
SLIDE 6

In

  • Introduction. LSDA (I)

(I)

slide-7
SLIDE 7

In

  • Introduction. LSDA (I)

(I)

7

slide-8
SLIDE 8

In

  • Introduction. LSDA (I)

(I)

8

Least-Squares Disclosure Attack

slide-9
SLIDE 9

In

  • Introduction. LSDA (II

(II)

9

slide-10
SLIDE 10

Real data:

10

220.000 emails sent between employees of Enron Corporation. http://www.cs.cmu.edu/~./enron/ 400.000 location check-ins from Gowalla social networking website. http://snap.stanford.edu/data/loc-gowalla.html 180.000 posts to the public mailing lists of Indimedia http://lists.indimedia.org/

slide-11
SLIDE 11

Motivation: LSDA’s analysis falls short

11

slide-12
SLIDE 12

Real-World Behavior. In Input process.

12

Emails dataset

slide-13
SLIDE 13

Real-World Behavior. Output process.

#messages Email Location Mailing List =2 1.85 1.03 1.29 =3 2.71 1.05 1.46 =4 3.53 1.06 1.53 =5 4.40 1.08 1.53 ≥6 13.56 1.11 1.57

13

Average number of receivers chosen

slide-14
SLIDE 14

New Theoretical Analysis (I) (I)

14

Multinomial output Maximum variance output “uniformity” Number of rounds observed Variance of the number

  • f messages sent

Global contribution Individual term

slide-15
SLIDE 15

New Theoretical Analysis (II (II)

15

slide-16
SLIDE 16

Results

16

slide-17
SLIDE 17

17

Thank you!!

simonoya@gts.uvigo.es