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
We are opening this post for feedback on the Reach Whitepaper. The whitepaper aims to provide ad tech companies with actionable guidance on implementing reach measurement within the Privacy Sandbox, via the Shared Storage and Private Aggregation APIs.
The methods that are covered include:
Direct Method
Point Contribution Method
Sketch Based Method
Our experiments show that for predetermined queries, the best solution is to use the point contribution mechanism. However for more exploratory queries or those with longer windows, sketch-based mechanisms are beneficial although techniques such as capping should be selected carefully to reduce error.
For the Sketch Based Method we comment that the decrease in utility is due to the addition of noise to each intermediate sketch before the final result is computed. We would be interested to hear feedback on whether a sketch based method is preferred if the API could evolve to allow ad techs to construct encrypted un-noised intermediate aggregations (i.e sketches) that can later be combined to produce a reach count, with noise only added to the final result.
The text was updated successfully, but these errors were encountered:
We are opening this post for feedback on the Reach Whitepaper. The whitepaper aims to provide ad tech companies with actionable guidance on implementing reach measurement within the Privacy Sandbox, via the Shared Storage and Private Aggregation APIs.
The methods that are covered include:
Direct Method
Point Contribution Method
Sketch Based Method
Our experiments show that for predetermined queries, the best solution is to use the point contribution mechanism. However for more exploratory queries or those with longer windows, sketch-based mechanisms are beneficial although techniques such as capping should be selected carefully to reduce error.
For the Sketch Based Method we comment that the decrease in utility is due to the addition of noise to each intermediate sketch before the final result is computed. We would be interested to hear feedback on whether a sketch based method is preferred if the API could evolve to allow ad techs to construct encrypted un-noised intermediate aggregations (i.e sketches) that can later be combined to produce a reach count, with noise only added to the final result.
The text was updated successfully, but these errors were encountered: