IPA Fake Event Generation

IPA has a number of records $N$. Each record consists of a secret shared match key and other associated data.

The task is to generate $M$ fake events that consists of secret shared random match keys (fake match key) and associated data set to $0$ for a given bound $B$.

More concretely: We need to do the following process for all combinations of two out of three helper parties in $(H_1,H_2,H_3)$:

the two selected helper parties, i.e. $(H_i,H_j)\in(H_1,H_2,H_3)$ do the following:

for each $m \in [1,...,B]$ do the following:
- Sample $e_m$ from a Gaussian distribution that accomplishes $(\epsilon, \delta)$-Differential Privacy for sensitivity $1$.
- for each $i\in [1,...,e_m]$ do the following: generate shares of a random fake match key (using prss), and generate $m$ events with associated data $0$ for this match key
compute $$M{i,j} = \sum{k=1}^n (m\cdot e_m)$$ and send it to the other helper party $H_k$, i.e. party not in $(H_i,H_j)$.

$Hk$ generates $M{i,j}$ fake events and sets all shares to $0$.

Append all $M:=M{1,2}+M{2,3}+M_{3,1}$ fake events to the initial $N$ records.

private-attribution / ipa

IPA Fake Event Generation #851