ICP5 (https://github.com/Hiresh12/UMKC/tree/master/CSEE5590%20-%20AI%20Cyber%20Security/ICP5/Source)

Global differential privacy

In LDP we are adding randomness to the data, but in global differential privacy, some noise will be added to the result of the aggregated function. That it ensures the accuracy of the results are maintained well than LDP.

The amount of noise to be added to the result is to be decided based on the epsilon value, Which in-turn will be decided by the sensitivity of the function over the data and by the laplace method

We will see in implementation,

Creating DB and parallel DB , and defining the sum function as query,

Function to calculate the sensitivity of the function over the data,

Function to add noise to the query,

We will see now how to calculate the noise, The below function is used to calculate the noise,

We will divide the sensitivity with epsilon to calculate the beta value, which is used to calculate the noise using laplace distribution. Below are the output of the sum and mean function for some of the epsilon values,