Re: [Model-t] Minimization

[Christian Huitema <[e-mail]>]

On 3/17/2022 11:12 AM, Mirja Kuehlewind wrote:
> I agree these are good points!
>
> I also just had a quick read and to me it seem that there might also be a point about identity. Often information is only sensitive if it can be associated to a certain identify, so providing data in an anonymous way whenever possible might also be a good approach. Not sure if this belongs in this document or if that is a separate point….


Yes, but there are also many examples of allegedly anonymized datasets 
that were then shown to contain enough information to identify the 
subjects. If we want to say something like that, then we need to be 
rather cautious, and distinguish between statistics and hidden 
identities. Saying "out of 1,000,000 connections, we observed that 10% 
had a delay larger than 100 ms" is one thing. Having data on individual 
sessions, even without explicitly storing PII, is another.

The flip side of that is of course debugging. It is hard to debug a 
complex failure without extensive data, such as a trace of all packets 
in a connection...

-- Christian Huitema