Transformed partial convolution algorithm for composition determination
Differential privacy composition determination in a secure communication is provided. A method for determining differential privacy composition includes determining a differential privacy configuration including a first privacy parameter and a second privacy parameter, determining a privacy loss distribution, and performing a partial convolution operation of the privacy loss distribution by transforming the privacy loss distribution based on a predetermined parameter, determining an integral range based on the first privacy parameter and the privacy loss distribution, transforming the privacy loss distribution based on the integral range, and performing the partial convolution operation based on the integral range. The method also includes determining the second privacy parameter based on a result of the partial convolution operation, and controlling a dataset based on the differential privacy configuration to limit access to the dataset.
1. A method for determining differential privacy composition in a secure communication, the method comprising:
determining a differential privacy configuration including a first privacy parameter and a second privacy parameter;
determining a privacy loss distribution;
performing a partial convolution operation of the privacy loss distribution by:
transforming the privacy loss distribution based on a predetermined parameter;
determining an integral range based on the first privacy parameter and the privacy loss distribution;
transforming the privacy loss distribution based on the integral range; and
performing the partial convolution operation based on the integral range;
determining the second privacy parameter based on a result of the partial convolution operation; and
controlling a dataset based on the differential privacy configuration to limit access to the dataset.
2. The method of claim 1 , wherein the transforming of the privacy loss distribution based on the predetermined parameter includes translating the privacy loss distribution.
3. The method of claim 2 , wherein the transforming of the privacy loss distribution based on the predetermined parameter further includes after the translating of the privacy loss distribution, reflecting the privacy loss distribution.
4. The method of claim 1 , wherein the transforming of the privacy loss distribution based on the integral range includes truncating the privacy loss distribution based on the integral range.
5. The method of claim 1 , wherein K is a number of convolution operations, K is greater than two, the performing of the partial convolution operation of the privacy loss distribution includes performing a K-fold partial convolution operation of the privacy loss distribution.
6. The method of claim 5 , wherein the determining of the integral range includes determining the integral range based on the first privacy parameter, the number of convolution operations K, and the privacy loss distribution.
7. The method of claim 1 , further comprising:
adjusting a number of composition operations based on the first privacy parameter and the second privacy parameter.
8. The method of claim 7 , wherein the adjusting of the number of composition operations includes:
decreasing the number of composition operations when the first or second privacy parameter is greater than a first predetermined threshold; and
increasing the number of composition operations when the first or second privacy parameter is less than a second predetermined threshold.
9. A differential privacy composition determination system comprising:
a memory to store a dataset; and
a processor to:
determine a differential privacy configuration including a first privacy parameter and a second privacy parameter;
determine a privacy loss distribution;
perform a partial convolution operation of the privacy loss distribution by:
transforming the privacy loss distribution based on a predetermined parameter;
determining an integral range based on the first privacy parameter and the privacy loss distribution;
transforming the privacy loss distribution based on the integral range; and
performing the partial convolution operation based on the integral range;
determine the second privacy parameter based on a result of the partial convolution operation; and
control the dataset based on the differential privacy configuration to limit access to the dataset.
10. The system of claim 9 , wherein the transforming of the privacy loss distribution based on the predetermined parameter includes translating the privacy loss distribution.
11. The system of claim 10 , wherein the transforming of the privacy loss distribution based on the predetermined parameter further includes after the translating of the privacy loss distribution, reflecting the privacy loss distribution.
12. The system of claim 9 , wherein the transforming of the privacy loss distribution based on the integral range includes truncating the privacy loss distribution based on the integral range.
13. A non-transitory computer-readable medium having computer-executable instructions stored thereon that, upon execution, cause one or more processors to perform operations comprising:
determining a differential privacy configuration including a first privacy parameter and a second privacy parameter;
determining a privacy loss distribution;
performing a partial convolution operation of the privacy loss distribution by:
transforming the privacy loss distribution based on a predetermined parameter;
determining an integral range based on the first privacy parameter and the privacy loss distribution;
transforming the privacy loss distribution based on the integral range; and
performing the partial convolution operation based on the integral range;
determining the second privacy parameter based on a result of the partial convolution operation; and
controlling a dataset based on the differential privacy configuration to limit access to the dataset.
14. The computer-readable medium of claim 13 , wherein the transforming of the privacy loss distribution based on the predetermined parameter includes translating the privacy loss distribution.
15. The computer-readable medium of claim 14 , wherein the transforming of the privacy loss distribution based on the predetermined parameter further includes after the translating of the privacy loss distribution, reflecting the privacy loss distribution.
16. The computer-readable medium of claim 13 , wherein the transforming of the privacy loss distribution based on the integral range includes truncating the privacy loss distribution based on the integral range.
17. The computer-readable medium of claim 13 , wherein K is a number of convolution operations, K is greater than two, the performing of the partial convolution operation of the privacy loss distribution includes performing a K-fold partial convolution operation of the privacy loss distribution.
18. The computer-readable medium of claim 17 , wherein the determining of the integral range includes determining the integral range based on the first privacy parameter, the number of convolution operations K, and the privacy loss distribution.
19. The computer-readable medium of claim 13 , the operations further comprise:
adjusting a number of composition operations based on the first privacy parameter and the second privacy parameter.
20. The computer-readable medium of claim 19 , wherein the adjusting of the number of composition operations includes:
decreasing the number of composition operations when the first or second privacy parameter is greater than a first predetermined threshold; and
increasing the number of composition operations when the first or second privacy parameter is less than a second predetermined threshold.