Get desktop application:
View/edit binary Protocol Buffers messages
Used in:
Distribution of the noising algo.
Required.
Privacy loss parameter. Inversely proportional to amount of noise added.
Required.
Sensitivity for noising aggregation count.
Sensitivity for noising aggregation value.
Used in:
Used in:
Maximum sum of contributions a single privacy unit can make for all partitions (keys). Two values with difference <= l1_sensitivity that are noised should produce roughly the same distribution. Aka global sensitivity
Config for privacy parameters
delta must be > 0 and < 1
epsilon must be > 0
l1_sensitivity must be > 0
The noise distribution should use the following parameters scale = l1_sensitivity / epsilon threshold = scale * (epsilon + ln(1/delta))