Closed rillian closed 4 months ago
cc @AliShahin @DJAndries
The above requires 1.68.101
or higher for 1.68.x
verification 👍
Verification PASSED on
Brave | 1.68.101 Chromium: 126.0.6478.126 (Official Build) beta (64-bit)
-- | --
Revision | ffa8411f4c8fe36c0e399f60ef7c24b3cf7d7402
OS | Windows 10 Version 22H2 (Build 19045.4529)
Verification PASSED on Vivo X70 Pro version12
running Bravemonoarm64.apk_1.68.101
star-randsrv.bsg.brave.com
and collector.bsg.brave.com
endpoints as usual
Description
Per recent discussions, we want to add differential privacy to the STAR protocol used to report how features are used in the browser. The first phase of this is just to add sampling to the current reports: some percentage of the time the client submits a measurement with the value or tag of the outer layer replaced with some random value. This addresses some of the leakage from undecodable tags in STAR by adding plausible deniability to the threshold anonymity.
Proposed parameters, derived from differential privacy ε=1 and δ = 10^-8:
By introducing sampling we enforce better privacy bounds even with the lower aggregation threshold (currently K=50).
Rough plan
Brave-Aggregation-Threshold
http header can mark the difference.