Timezone: »
Computing Differential Privacy Guarantees for Heterogeneous Compositions Using FFT
Antti Koskela · Antti Honkela
The recently proposed Fast Fourier Transform (FFT)-based accountant for evaluating $(\varepsilon,\delta)$-differential privacy guarantees using the privacy loss distribution formalism has been shown to give tighter bounds than commonly used methods such as Rényi accountants when applied to homogeneous compositions, i.e., to compositions of identical mechanisms. In this work, we extend this approach to heterogeneous compositions. We carry out a full error analysis that allows choosing the parameters of the algorithm such that a desired accuracy is obtained. The analysis also extends previous results by taking into account all the parameters of the algorithm. We also show how to speed up the evaluation of tight privacy guarantees using the Plancherel theorem at the cost of increased pre-computation and memory usage.
Author Information
Antti Koskela (University of Helsinki)
Antti Honkela (University of Helsinki)
More from the Same Authors
-
2021 : Tight Accounting in the Shuffle Model of Differential Privacy »
Antti Koskela · Mikko A Heikkilä · Antti Honkela -
2021 : Differentially Private Hamiltonian Monte Carlo »
Ossi Räisä · Antti Koskela · Antti Honkela -
2021 : Gaussian Processes with Differential Privacy »
Antti Honkela -
2021 Poster: Differentially Private Bayesian Inference for Generalized Linear Models »
Tejas Kulkarni · Joonas Jälkö · Antti Koskela · Samuel Kaski · Antti Honkela -
2021 Spotlight: Differentially Private Bayesian Inference for Generalized Linear Models »
Tejas Kulkarni · Joonas Jälkö · Antti Koskela · Samuel Kaski · Antti Honkela -
2017 Workshop: Private and Secure Machine Learning »
Antti Honkela · Kana Shimizu · Samuel Kaski