Andrew Pascoe

As far as NextRoll is concerned, we greatly prefer UTC in all of our logs, with any translations (say for a customer) happening downstream.

I know what you mean, but that's not what I meant. We would just take the UTC timestamp of the bucketed day and translate that for the customer so they...

We had some long discussions at NextRoll about this. I think we're still gathering our thoughts for a more thorough response, but here's a quick brief of our initial thoughts....

Hi, Michael, 1) I understand the tracking attack, but it is unclear to me how the ML features would be sent to the malicious tracker. If you're using the MURRE...

At first glance, that doesn't appear to be sufficient to me. 1) The L1-sensitivity is a function of the _function_ that is run over the data. The size of the...

Yes, summing is a very direct computation for L1-sensitivity, but if you're doing that, you may as well scale the noise with the maximum input. There's no need to limit...

Is that true? Unless you also report back the distribution of the noise, I don't think you can make that claim. The receiver of the report, in your $1B house...

I _think_ that's a misapplication of differential privacy. You shouldn't have two separate databases, but rather treat them as a single large database. In that case, the L1-sensitivity should be...