Skip to main content
. 2020 Nov 3;22(11):e18735. doi: 10.2196/18735

Figure 3.

Figure 3

Applying obfuscation to HLL sketches. (a) HLL+Mask: For each bucket, we count the total number of patients (not just the ones who match the query) whose hashes have the same leading 1-indicator. If that number is less than 10, then the bucket is not 10-anonymous, so we do not send the HLL sketch. Instead, we only send a masked aggregate count of the number of patients matching the query. (b) HLL+Shuffle: We do a coordinated random shuffling so the central hub does not know what the original buckets were for the leading 1 indicator. However, the hub can still estimate cardinality in the same way as HLL without obfuscation. HLL: HyperLogLog.