Table 7 The five privacy metrics on Ship-D and California House datasets.
From: A tabular data generation framework guided by downstream tasks optimization
Datasets | Method | kAnonymization | lDiversityDistinct | kMap | DeltaPresence | IdentifiabilityScore | |||
|---|---|---|---|---|---|---|---|---|---|
gt | syn | gt | syn | Score | Score | Score | socre_OC | ||
Ship-D | RTVAE | 999 | 43 | 999 | 43 | 31 | 42.52 | 0.0232 | 0.0298 |
CTGAN | 43 | 39 | 43 | 39 | 22 | 2.36 | 0.3710 | 0.3490 | |
TabDDPM | 3 | 6 | 3 | 6 | 3 | 3.33 | 0.5400 | 0.0400 | |
DDPM with classifier | 3 | 2 | 3 | 2 | 1 | 9.00 | 0.3200 | 0.0500 | |
ShipGen | 3 | 4 | 3 | 4 | 2 | 5.00 | 0.4100 | 0.0300 | |
TDGGD | 3 | 2 | 3 | 2 | 1 | 8.00 | 0.4200 | 0.0100 | |
California House | RTVAE | 1 | 3 | 1 | 3 | 5 | 3.20 | 0.0230 | 0.1180 |
CTGAN | 1 | 1 | 1 | 1 | 3 | 3.72 | 0.3190 | 0.1180 | |
TabDDPM | 1 | 1 | 1 | 1 | 4 | 3.17 | 0.3200 | 0.0600 | |
DDPM with classifier | 1 | 3 | 1 | 3 | 2 | 9.50 | 0.1900 | 0.0500 | |
ShipGen | 1 | 1 | 1 | 1 | 1 | 2.60 | 0.1900 | 0.0300 | |
TDGGD | 1 | 1 | 1 | 1 | 1 | 0.90 | 0.0100 | 0.0010 | |
