Ted

Agreed, thanks for the fast answer and for looking at this! I think your proposed solution is a good idea. Two more comments on the same topic: - I'm not...

Do @BenjaminDev's patches solve this error, @Dipinka, or is the workaround still necessary?

Small addendum: my initial sentence that "when summing two floating-point numbers, the result has the precision of the least-precise summand" is technically incorrect. It should be "when summing two floating-point...

As another small addendum, so there's a trace of this somewhere: this also affects OpenDP's Gaussian and Analytic Gaussian mechanisms, as well as all floating-point noise primitives in SmartNoise (Laplace,...

Hi folks, I am wondering whether there was any progress on this issue. The latest OpenDP newsletter mentions improvements on "the implementation details [...] particularly around the handling of floating...

Yes, discretizing the output space and generating a two-sided geometric instead of a Laplace distribution is [what Google DP does](https://github.com/google/differential-privacy/blob/main/common_docs/Secure_Noise_Generation.pdf) (see Section 4), and I think is a valid solution...

You're correct that the integerization approach requires setting a granularity, and that once you have this granularity, if you don't do any clamping, you technically get a vulnerability. For example,...

Great question! I like this video a lot, and I think alternative materials would still be super valuable. In particular: - The video focuses a lot on explaining database reconstruction...

Happy to try this out as well!

Thanks @naoise-h. I'm looking forward to read about your new approach! One follow-up question: do you intend to mention these vulnerabilities, and explicitly point out that diffprivlib should not be...