Skip to content
This repository has been archived by the owner on Sep 21, 2022. It is now read-only.

Derive constants c1 and c2 #314

Open
ecowan opened this issue Nov 2, 2020 · 2 comments
Open

Derive constants c1 and c2 #314

ecowan opened this issue Nov 2, 2020 · 2 comments
Labels

Comments

@ecowan
Copy link
Contributor

ecowan commented Nov 2, 2020

By Abadi et al. theorem 1 (pg. 4) we can choose constants c1 and c2 such that sigma will guarantee differential privacy

@ecowan ecowan added the CSL label Nov 2, 2020
@ecowan
Copy link
Contributor Author

ecowan commented Nov 4, 2020

From Theorem 1 (pg. 14) we have a DP guarantee provided that

  1. T * q^2 * lambda^2 / sigma^2 <= lamba * epsilon / 2
  2. exp(-lambda * epsilon / 2) <= delta

So given epsilon and delta from the inputs to the function, we need to find the optimal sigma (noise scale) to satisfy (1) and also verify that (2) is satisfied.

@ecowan
Copy link
Contributor Author

ecowan commented Nov 4, 2020

I believe this can be merged in with #313

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
Projects
None yet
Development

No branches or pull requests

1 participant