Module `Blip`

BLIP: BLoom-then-flIP

This is an implementation of the BLoom-then-flIP algorithm as described in the paper BLIP: Non-interactive Differentially-Private Similarity Computation on Bloom filters

val p : float -> int -> float

p e k returns the optimal flipping probability of a Bloom filter.

parameter e: Differential privacy parameter ε that determines the privacy-utility trade-off: the smaller e is, the more privacy, but also the less utility. e <= 10 is the upper bound that still prevents an adversary from performing a successful profile reconstruction attack.

parameter k: Number of hash functions used in the Bloom filter.

val flip : ?⁠g:Nocrypto.Rng.g -> Bitv.t -> float -> Bitv.t: flip ?g bits p flips each element of bits independently with probability p, using RNG g if provided

val sim : Bitv.t -> Bitv.t -> float: sim a b returns the cosine similarity measure of two bit vectors representing (flipped) Bloom filters