Locality-Sensitive Hashing Does Not Guarantee Privacy!
Attacks on Google's FLoC and the MinHash Hierarchy
Our work focuses on two popular functions from the locality-sensitive hashing (LSH) family, namely SimHash and MinHash. Both are used in Federated Learning of Cohorts (FLoC) and the MinHash Hierarchy system respectively.
This repository contains the code for the reproduction of privacy systems based on LSH and our attacks, breaking the privacy properties of those systems, along with their experimental evaluation on real datasets.
Since the LSH-based systems evaluated are independent, their implementation along with our attacks can be found in their respective folder: FLoC/ and MinHash_Hierarchy/.
More details about our attacks and their evaluation can be found in the paper "Locality-Sensitive Hashing Does Not Guarantee Privacy! Attacks on Google's FLoC and the MinHash Hierarchy".