CompositeDP

1. Background: This is a novel composite DP mechanism.

2. Install: There are 6 Python files in total:

    1) Perturbation_Mechanism.py
    
    2) Traditional_Mechanism.py
    
    3) Mechanism/Parameter_Optimization.py
    
    4) Mechanism/Constraints.py
    
    5) Mechanism/Evaluation.py
    
    6) Mechanism/Mapping.py
   

   For using this novel DP Library, users need to download all Python files in their local environment.

3. Usage: This is a Python library, users can just import it.

   Import library:

   from Traditional_Mechanism import *

   from Perturbation_Mechanism import *

    	Notes: For the library discretegauss and cdp2adp, please reference to https://github.com/IBM/discrete-gaussian-differential-privacy.
   

   Then users can use the composite DP mechanism freely.

4. Using examples: Please see the Example1.py file.

5. Perturbation Function List:

   Perturbation-1: Activation-1+Base-1, index=1

   Perturbation-2: Activation-2+Base-1, index=2

   Perturbation-3: Activation-3+Base-1, index=3

   Perturbation-4: Activation-1+Base-2, index=4

   Perturbation-5: Activation-2+Base-2, index=5

   Perturbation-6: Activation-3+Base-2, index=6

6. How to use perturbation function:

   1. For one time call:

      ep: Privacy Budget

      fd: Raw input query result f(D)

      sensitivity: Sensitivity value

      lower: The lower_bound of the query f

      k: parameter

      m: parameter

      y: parameter

      index: which perturbation function you are going to use

      Call Function: perturbation_fun_oneCall(ep, fd, sensitivity, lower, k, m, y, index)

      Return value: Perturbed Result Op

   2. For multiple call:

      ep: Privacy Budget

      fd: Raw input query result f(D)

      sensitivity: Sensitivity value

      lower: The lower_bound of the query f

      k: parameter

      m: parameter

      y: parameter

      index: which perturbation function you are going to use

      repeat_times: How many times you are going to recall the mechanism

      Call Function: perturbation_fun_multipleCall(ep, fd, sensitivity, lower, k, m, y, index, repeat_times)

      Return value: A list of perturbed Results Op_Multiple

   If you don't want to pick the parameters by yourself, you can use optimized function:

   3. For one time call (optimized):

      ep: Privacy Budget

      fd: Raw input query result f(D)

      sensitivity: Sensitivity value

      lower: The lower_bound of the query f

      index: which perturbation function you are going to use

      Call Function: perturbation_fun_optimized_oneCall(ep, fd, sensitivity, lower, index)

      Return value: Perturbed Result Op

   4. For multiple call (optimized):

      ep: Privacy Budget

      fd: Raw input query result f(D)

      sensitivity: Sensitivity value

      lower: The lower_bound of the query f

      index: which perturbation function you are going to use

      repeat_times: How many times you are going to recall the mechanism

      Call Function: perturbation_fun_optimized_multipleCall(ep, fd, sensitivity, lower, index, repeat_times)

      Return value: A list of perturbed Results Op_Multiple

Thanks very much!