You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Mar 16, 2023. It is now read-only.
@michaelkleber for a given k-anonymity of k, suppose you have a prefix 110101 which contains more than k users, and his brother 110100 contains less than k users, do you choose 110101 to be the floc, and drops all users whose hash is prefixed by 110100, or do you ride up to the node above it, choosing 11010 as floc (and not 110101), in order to cover more users even if it is less precise ?
I understand that flocs are not renumbered frequently in order to keep ads campaigns sustainable. But from what you say I also understand that, more frequently, number of users in each floc are recomputed, to drop flocs which happen to have too few users now. At which frequency are these nb of users by floc recomputed ?
Can a floc be eliminated one day and come back another day because it now contains enough users ?
Thanks @michaelkleber
The text was updated successfully, but these errors were encountered:
@michaelkleber for a given k-anonymity of k, suppose you have a prefix 110101 which contains more than k users, and his brother 110100 contains less than k users, do you choose 110101 to be the floc, and drops all users whose hash is prefixed by 110100, or do you ride up to the node above it, choosing 11010 as floc (and not 110101), in order to cover more users even if it is less precise ?
The second thing you said — we would use the prefix 11010 to define the cohort, rather than divide it into two pieces one of which is too small. I tried to explain PrefixLSH here, in case you haven't seen it.
I understand that flocs are not renumbered frequently in order to keep ads campaigns sustainable. But from what you say I also understand that, more frequently, number of users in each floc are recomputed, to drop flocs which happen to have too few users now. At which frequency are these nb of users by floc recomputed ?
We haven't come up with a refresh schedule for this. One part of the Origin Trial is that it will give us a chance to collect enough data to figure out a good answer to this question.
Can a floc be eliminated one day and come back another day because it now contains enough users ?
Yes, that's not impossible. It seems like it would be potentially disruptive to folks using flocks, so if it seems likely to happen often, we should probable build some hysteresis into the update system to avoid that kind of churn.
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Following this issue #44
Thanks @michaelkleber
The text was updated successfully, but these errors were encountered: