Results 1 - 10 of 68,085

Table 3. Susceptibility of Genetic Disease Populations to REIDIT-C Re-identification.

in How (Not) to Protect Genomic Data Privacy in a Distributed Network: Using Trail Re-identification to Evaluate and Design Anonymity Protection Systems
by Bradley Malin, Latanya Sweeney 2004
Cited by 12

Table 3: Re-identification experiments using dataset Census and method IPSO-C. Results in number of correct re-identifications over an overall number of 1080 records.

in Re-Identification and Synthetic Data Generators: A Case Study ∗† Abstract
by Josep Domingo-ferrer, Vicenç Torra, Josep M. Mateo-sanz, Francesc Sebé

Table 2. Classification of re-identifications made by REIDIT-I. Light-shaded cells are possible outcomes and the darkened cell is an impossible outcome.

in How (Not) to Protect Genomic Data Privacy in a Distributed Network: Using Trail Re-identification to Evaluate and Design Anonymity Protection Systems
by Bradley Malin, Latanya Sweeney 2004
"... In PAGE 8: ... This process continues until no more re-identifications can be made because one of two conditions is satisfied: either (1) the track with incomplete trails has no more trails to process; or, (2) there are no re-identifications made in the current iteration. REIDIT-I can generate the four possible results for two arbitrary trails trail(N,n) and trail(P,p), as shown in Table2 : 1) correct match, 2) correct non-match, 3) false non-match, and 4) false match. The first three can occur, while the last is impossible.... ..."
Cited by 12

Table 4. Matching Counts and Truth Probabilities By Total Income Category Re-Identification Pass, Masked/Swapped File

in Re-identification methods for evaluating the confidentiality of analytically valid microdata
by William E. Winkler 1998
"... In PAGE 10: ... The combination of swapping and additive-noise procedures used in creating the file used in the second pass has the advantage that easily re-identified records in the masked-only file are generally non-re-identifiable and that means and covariances are approximately preserved on the entire set of pairs and on important subdomains. We observe ( Table4 ) that the re-identification rate is effectively negligible in the file used in the second pass. Use of the additive noise procedure of Kim [13] allows us to recover means and correlations of important statistics.... In PAGE 12: ... In the last column, 5% of records with incomes below $80000 and all records with incomes above $80000 are swapped. The more complete set of swapping assures that the more easily identified large income individuals are not likely to be re-identified as is shown in Table4 . In Table 6, we show how correlations may not be preserved in the subdomain of records having some of their information taken from IRS Schedule C.... ..."
Cited by 19

Table 1. Classification of re-identifications made by REIDIT-C. Light-shaded cells are possible outcomes and the darkened cell is an impossible outcome.

in How (Not) to Protect Genomic Data Privacy in a Distributed Network: Using Trail Re-identification to Evaluate and Design Anonymity Protection Systems
by Bradley Malin, Latanya Sweeney 2004
"... In PAGE 7: ...ig. 3. Pseudocode for REIDIT-C. REIDIT-C can generate the four possible results for two arbitrary trails trail(N,n) and trail(P,p), as shown in Table1 : 1) correct match, 2) correct non-match, 3) false non-match, and 4) false match. The first three can occur, while the last is impossible.... ..."
Cited by 12

Table 1. Re-identification experiments using dataset Census and methods IPSO-A, IPSO-B and IPSO-C

in Using Mahalanobis Distance-Based Record Linkage for Disclosure Risk Assessment
by Vicenç Torra, John M. Abowd, Josep Domingo-ferrer 2006
"... In PAGE 6: ... Released files (see [9, 8] for details) were generated using the synthetic data generators IPSO-A, IPSO-B and IPSO-C. Table1 lists the sets of quasi-identifiers considered for the Census data in the case of data generated using IPSO-A. Analogous sets of quasi-identifiers (viS1 B and viS1 C instead of viS1 A ) were considered for the other IPSO-B and IPSO-C methods.... ..."
Cited by 1

Table 2. Re-identification experiments using dataset EIA and methods IPSO-A, IPSO-B and IPSO-C

in Using Mahalanobis Distance-Based Record Linkage for Disclosure Risk Assessment
by Vicenç Torra, John M. Abowd, Josep Domingo-ferrer 2006
"... In PAGE 6: ... Analogous sets of quasi-identifiers (viS1 B and viS1 C instead of viS1 A ) were considered for the other IPSO-B and IPSO-C methods. Table2 contains similar information corresponding to EIA datasets. Note that in this paper only experiments with files sharing attributes... ..."
Cited by 1

Table 5: Re-identification experiments using dataset EIA and methods IPSO-A, IPSO-B and IPSO-C. Results in number of correct re-identifications over an overall number of 4092 records.

in Re-Identification and Synthetic Data Generators: A Case Study ∗† Abstract
by Josep Domingo-ferrer, Vicenç Torra, Josep M. Mateo-sanz, Francesc Sebé
"... In PAGE 6: ...tributes the quasi-identifier attributes Y . The notation in Table5 below is the same used in the analogous tables for the Census dataset, except that no scenario superscript is used. The table shows the results of record linkage experi- ments between the EIA dataset and partially synthetic versions of it generated using IPSO-A, IPSO-B and IPSO-C.... In PAGE 6: ... Only the quasi-identifiers used in each experiment are listed, which are subsets of those specified in Table 4. Quasi-identifiers in Table5 were selected using the cross-correlation matrix between the orig- inal quasi-identifier attributes and the quasi- identifier attributes generated using methods IPSO-A, IPSO-B, IPSO-C. The rationale of our quasi-identifier choices is that at least some of the quasi-identifiers in datasets A and B should be highly correlated.... ..."

Table 1: Results of the re-identification. Case 1: Files with the following normal distributions N(0; 0:5)::N(4; 2); Case 2: Files with the following normal distributions N(0; 0:5)::N(8; 2)

in
by unknown authors
"... In PAGE 3: ....3. Results We have obtained good results with all fuzzy measures. The results are given in Table1 . The table contains the best number of re-identifications obtained for each experiment with either distance-based or probabilistic record linkage.... ..."

Table 6. Re-identification experiments using dataset EIA and methods IPSO- A, IPSO-B and IPSO-C. Results in number of correct re-identifications over an overall number of 4092 records. DBRL1: attribute-standardizing implementation of distance-based record linkage (DBRL); DBRL2: distance-standardizing implementa- tion of DBRL; DBRLM-COV and DBRLM-COV0: distance-based record linkage using Mahalanobis distance (covariances computed using the appropriate alingment or co- variances set to zero); KDBRL: distance-based record linkage with kernel distance (polynomic kernel with d=2); PRL: probabilistic record linkage

in Using Mahalanobis Distance-Based Record Linkage for Disclosure Risk Assessment
by Vicenç Torra, John M. Abowd, Josep Domingo-ferrer 2006
Cited by 1
Results 1 - 10 of 68,085