Skip to main content
5 votes
Accepted

Is the differential privacy definition with lower and upper bound equivalent to the definition with just an upper bound?

Yes, the definitions are equivalent. That the second implies the first is obvious. To see that the first implies the second, note that the datasets $D_1,D_2$ can be interchanged. So, if $$ \frac{\...
Discrete lizard's user avatar
  • 8,303
2 votes
Accepted

Histogram representation of a database

No, you went wrong right from the start. $x$ is the database. In this case, if the database is shown in that table, $x$ is the set of all rows shown in the table above. If $x$ is a database with ...
D.W.'s user avatar
  • 162k
1 vote
Accepted

What is the sensitivity of gradient clipping in Differentially Private SGD?

The notion of sensitivity is relative to the definition of adjacency. Sensitivity is the maximum of the absolute distance $|f(d) − f(d)|$ where $d$ and $d'$ are adjacent inputs. Adjacency may be ...
Nicolas Grislain's user avatar
1 vote

Differential Privacy with only positive noise

what you are looking for is the "exponential mechanism". See https://www.cis.upenn.edu/~aaroth/Papers/privacybook.pdf. It is a very general mechanism, which includes one-sided noise as a ...
Luigi Scorzato's user avatar

Only top scored, non community-wiki answers of a minimum length are eligible