search

IBM / differential-privacy-library

Diffprivlib: The IBM Differential Privacy Library
https://diffprivlib.readthedocs.io
MIT License
834 stars 200 forks source link

[question] About the gamma distribution noise used in the vector mechanism #56

Closed arafubeatbox closed 3 years ago

arafubeatbox commented 3 years ago

Hello and thank you for providing such a wonderful privacy preserving library.

It seems that the gamma distribution is used for producing noise to add for the vector mechanism, but reading the paper of Differentially Private Empirical Risk Minimization I could not find such logic. Is there a proof in past papers for gamma distribution noise providing differential privacy?

I looked at other issues posting questions, and thought this was the best way to ask a question you guys. Excuse me if I am misusing the github issues for this project, and thank you in advance.

arafubeatbox commented 3 years ago

@naoise-h Any chance I can get a response to this question?

arafubeatbox commented 3 years ago

I've found the paper I was looking for.

Chaudhuri, Kamalika, and Claire Monteleoni. "Privacy-preserving logistic regression." NIPS. Vol. 8. 2008.