Sarve: synthetic data and local differential privacy for private frequency estimation

Cybersecurity

Table 3 The lowest MSE for the existing method and the enhancements which were tested on the UCI adult dataset

Method	ε = ln(2)	ε = ln(3)	ε = ln(4)	ε = ln(5)	ε = ln(6)	ε = ln(7)
RS + FD[ADP]	0.000596388	0.000325887	0.000278437	0.000183621	0.000162579	0.000126356
RAPPOR	0.000821897	0.000474918	0.000429617	0.000391613	0.000293406	0.000274562
Hadamard response	0.001347787	0.000483257	0.000394622	0.00029166	0.00023901	0.000161183
Sarve	0.000559558	0.000315456	0.000243588	0.000190493	0.000150871	0.00014343

Method	ε = 2	ε = 3	ε = 4	ε = 5	ε = 6	ε = 7
RS + FD[ADP]	0.000132441	0.000124503	3.01E−05	2.16E−05	1.39E−05	1.97E−05
RAPPOR	0.0002658	0.00013092	7.36E−05	5.38E−05	4.08E−05	2.55E−05
Hadamard response	0.000198588	0.000135697	0.000103984	9.55E−05	0.000137056	8.82E−05
Sarve	0.000111824	5.53E−05	3.16E−05	2.19E−05	1.54E−05	1.60E−05

The values in bold indicate privacy conditions when Sarve performed better than adaptive RS+FD and resulted in lower MSE between real and post-privatization estimated frequencies