1. Dilys Thomas PODS 20061 Achieving Anonymity via Clustering G. Aggarwal, T. Feder, K. Kenthapadi, S. Khuller, R. Panigrahy, D. Thomas, A. Zhu

2. Dilys Thomas PODS 20062 Talk outline k-Anonymity model Achieving Anonymity via Clustering r-Gather clustering Cellular clustering Future Work

3. Dilys Thomas PODS 20063 Medical Records IdentifyingSensitive SSNNameDOBRaceZip codeDisease 614Sara03/04/76Cauc94305Flu 615Joan07/11/80Cauc94307Cold 629Kelly05/09/55Cauc94301Diabetes 710Mike11/23/62Afr-A94305Flu 840Carl11/23/62Afr-A94059Arthritis 780Joe01/07/50Hisp94042Heart problem 619Rob04/08/43Hisp94042Arthritis

4. Dilys Thomas PODS 20064 De-identified Medical Records Sensitive AgeRaceZip codeDisease Cauc94305Flu 07/11/80Cauc94307Cold 05/09/55Cauc94301Diabetes 11/23/62Afr-A94305Flu 11/23/62Afr-A94059Arthritis 01/07/50Hisp94042Heart problem 04/08/43Hisp94042Arthritis 03/04/76

5. Dilys Thomas PODS 20065 k-Anonymity model Uniquely identify you! Sensitive DOBRaceZip codeDisease 03/04/76Cauc94305Flu 07/11/80Cauc94307Cold 05/09/55Cauc94301Diabetes 12/30/72Afr-A94305Flu 11/23/62Afr-A94059Arthritis 01/07/50Hisp94042Heart problem 04/08/43Hisp94042Arthritis Quasi-identifiers: approximate foreign keys

6. Dilys Thomas PODS 20066 k-Anonymity Model [Swe00] Suppress some entries of quasi-identifiers – each modified row becomes identical to at least k-1 other rows with respect to quasi-identifiers Individual records hidden in a crowd of size k

7. Dilys Thomas PODS 20067 2-Anonymized Table DOBRaceZip codeDisease *Cauc*Flu *Cauc*Cold *Cauc*Diabetes 11/23/62Afr-A*Flu 11/23/62Afr-A*Arthritis *Hisp94042Heart problem *Hisp94042Arthritis

8. Dilys Thomas PODS 20068 k-Anonymity Optimization Minimize the number of generalizations/ suppressions to achieve k-Anonymity NP-hard to come up with minimum suppressions/ generalizations.[MW04]  (k) approximation for k-anonymity [AFK+05]  (k) lower bound on approximation ratio with graph assumption

9. Dilys Thomas PODS 20069 Talk outline k-Anonymity model Achieving Anonymity via Clustering r-Gather clustering Cellular Clustering Future Work

10. Dilys Thomas PODS 200610 Original Table AgeSalary Amy2550 Brian2760 Carol29100 David35110 Evelyn39120

11. Dilys Thomas PODS 200611 2-Anonymity with Suppression AgeSalary Amy** Brian** Carol** David** Evelyn** All attributes suppressed

12. Dilys Thomas PODS 200612 Original Table AgeSalary Amy2550 Brian2760 Carol29100 David35110 Evelyn39120

13. Dilys Thomas PODS 200613 2-Anonymity with Generalization AgeSalary Amy20-3050-100 Brian20-3050-100 Carol20-3050-100 David30-40100-150 Evelyn30-40100-150 Generalization allows pre-specified ranges

14. Dilys Thomas PODS 200614 Original Table AgeSalary Amy2550 Brian2760 Carol29100 David35110 Evelyn39120

15. Dilys Thomas PODS 200615 2-Anonymity with Clustering AgeSalary Amy[25-29][50-100] Brian[25-29][50-100] Carol[25-29][50-100] David[35-39][110-120] Evelyn[35-39][110-120] Cluster centers published 27=(25+27+29)/3 70=(50+60+100)/3 37=(35+39)/2 115=(110+120)/2

16. Dilys Thomas PODS 200616 Advantages of Clustering Clustering reduces the amount of distortion introduced as compared to suppressions / generalizations Clustering allows constant factor approximation algorithms

17. Dilys Thomas PODS 200617 Quasi-Identifiers form a Metric Space Convert quasi-identifiers into points in a metric space Distance function, D, on points –D(X,X)=0 Reflexive –D(X,Y)=D(Y,X) Symmetric –D(X,Z) <= D(X,Y) + D(Y,Z) Triangle Inequality

18. Dilys Thomas PODS 200618 Metric Space Converting (gender, zip code, DOB) into points in a metric space not easy. Define distance function on each attribute. E.g. on Zip code: –D (Zip1,Zip2)= physical distance between locations Zip1 and Zip2. Weight attributes, weighted sum of attribute distances gives metric.

19. Dilys Thomas PODS 200619 Clustering for Anonymity Cluster Quasi-identifiers so that each cluster has at least r members for anonymity. Publish cluster centers for anonymity with number of point and radius Tight clusters  Usefulness of data for mining Large number of points per cluster  Anonymity

20. Dilys Thomas PODS 200620 Quasi-identifiers: Metric Space Assume further that the distance metric has been already defined on quasi-identifiers

21. Dilys Thomas PODS 200621 Talk outline k-Anonymity model Achieving Anonymity via Clustering r-Gather clustering Cellular Clustering Future Work

22. Dilys Thomas PODS 200622 r-Gather Clustering 10 points, radius 5 20 points, radius 10 50 points, radius 20 Minimize the maximum radius: 20

23. Dilys Thomas PODS 200623 Results 2 Approximation to minimize maximum radius with cluster size constraint Matching Lower bound of 2 for maximum radius minimization

24. Dilys Thomas PODS 200624 r-Gather Clustering 2d

25. Dilys Thomas PODS 200625 Lower Bound: Reduction from 3-SAT X1TX1T X1FX1F X2TX2T X2FX2F r-2 points r-gather with radius 1 iff formula satisfiable Else radius ¸ 2 C 1 =X 1 Æ X 2 C1C1

26. Dilys Thomas PODS 200626 Talk outline k-Anonymity model Achieving Anonymity via Clustering r-Gather clustering Cellular Clustering Future Work

27. Dilys Thomas PODS 200627 Cellular Clustering 10 points, radius 5 20 points, radius 10 50 points, radius 20

28. Dilys Thomas PODS 200628 Cellular Clustering Metric 10 points, radius 5 20 points, radius 10 50 points, radius 20 Cellular Clustering Metric: 10*5 + 20*10 + 50*20 = 50 + 200 + 1000 = 1250

29. Dilys Thomas PODS 200629 Cellular Clustering Primal dual 4-approximation algorithm for cellular clustering Constant factor approximation to minimum cluster size –Each cluster has at least r points

30. Dilys Thomas PODS 200630 Cellular Clustering: Linear Program Minimize  c (  i x ic d c + f c y c ) Sum of Cellular cost and facility cost Subject to:  c x ic ¸ 1 Each Point belongs to a cluster x ic · y c Cluster must be opened for point to belong 0 · x ic · 1 Points belong to clusters positively 0 · y c · 1 Clusters are opened positively

31. Dilys Thomas PODS 200631 Dual Program Maximize  i  i Subject to:  i  ic · f c (1)  i -  ic · d c (2)  i ¸ 0  ic ¸ 0 Overview of Algorithm: First grow  i keeping  ic =0 till (2) becomes tight then grow  ic at same rate till (1) becomes tight

32. Dilys Thomas PODS 200632 Future Work Improve approximation ratio for Cellular Clustering Improve Running time. Presently r-gather is O(n 2 ) while cellular clustering is a linear program over n 2 variables. –Linear or even sub-linear time algorithms Weaker guarantees on anonymity, e.g. at least k/2 points per cluster instead of k.

33. Dilys Thomas PODS 200633 THANK YOU! QUESTIONS?