Computing Tight Differential Privacy Guarantees Using FFT

Show simple item record

dc.contributor Helsingin yliopisto, Tietojenkäsittelytieteen osasto fi
dc.contributor Helsingin yliopisto, Antti Honkela / Vastuullinen tutkija fi
dc.contributor.author Koskela, Antti
dc.contributor.author Honkela, Antti
dc.contributor.author Jälkö, Joonas
dc.date.accessioned 2020-06-30T11:41:01Z
dc.date.available 2020-06-30T11:41:01Z
dc.date.issued 2019-11-11
dc.identifier.citation Koskela , A , Honkela , A & Jälkö , J 2019 , ' Computing Tight Differential Privacy Guarantees Using FFT ' , TPDP 2019 - Theory and Practice of Differential Privacy CCS Workshop , Lontoo , Britannia , 11/11/2019 . en
dc.identifier.citation conference en
dc.identifier.other PURE: 139825457
dc.identifier.other PURE UUID: c1c4a958-b697-4005-8eda-03a4f3c2acd9
dc.identifier.other ORCID: /0000-0001-9193-8093/work/76725565
dc.identifier.uri http://hdl.handle.net/10138/317154
dc.description.abstract Computing privacy parameters for the differentially private stochastic gradient descent method (DP-SGD) is equivalent to analysing one dimensional mechanisms. We propose a numerical accountant for evaluating the (ε, δ)-privacy loss for mech- anisms with continuous one dimensional output. The proposed method is based on a numerical approximation of an integral formula which gives the tight (ε, δ)-values. The approximation is carried out by discretising the integral and by evaluating the resulting discrete convolutions using the fast Fourier transform algorithm. We focus on the subsampled Gaussian mechanism which underlies DP-SGD. We give both theoretical error bounds and numerical error estimates for the approximation. Experimen- tal comparisons with state-of-the-art techniques demonstrate significant improvements in bound tightness and/or computation time. Python code for the method can be found in Github (https://github.com/DPBayes/PLD-Accountant/). fi
dc.language.iso fin
dc.rights en
dc.subject 113 Tietojenkäsittely- ja informaatiotieteet en
dc.title Computing Tight Differential Privacy Guarantees Using FFT fi
dc.type Posteri
dc.type.uri info:eu-repo/semantics/other
dc.contributor.pbl
dc.contributor.pbl

Files in this item

Total number of downloads: Loading...

Files Size Format View
poster_computing_tight_DP_2.pdf 199.2Kb PDF View/Open

This item appears in the following Collection(s)

Show simple item record