Cryptology ePrint Archive: Report 2017/1234
High-Precision Privacy-Preserving Real-Valued Function Evaluation
Christina Boura and Ilaria Chillotti and Nicolas Gama and Dimitar Jetchev and Stanislav Peceny and Alexander Petric
Abstract: We propose a novel multi-party computation protocol for evaluating
continuous real-valued functions with high numerical precision.
Our method is based on approximations with Fourier series and uses
at most two rounds of communication during the online phase.
For the offline phase, we propose a trusted-dealer and honest-but-curious aided solution, respectively.
We apply our algorithm to train a logistic regression classifier via
a variant of Newton's method (known as IRLS) to compute unbalanced classification problems that detect rare events
and cannot be solved using previously proposed privacy-preserving
optimization algorithms (e.g., based on piecewise-linear approximations
of the sigmoid function).
Our protocol is efficient as it can be implemented using standard quadruple-precision floating point arithmetic. We report multiple experiments and provide a demo application that implements our algorithm for training a logistic regression model.
Category / Keywords: cryptographic protocols / Multiparty Computation, Numerical Precision, Fourier, Machine Learning
Original Publication (with minor differences): Financial Crypto 2017
Date: received 22 Dec 2017
Contact author: nicolas at inpher io
Available format(s): PDF | BibTeX Citation
Version: 20171222:210415 (All versions of this report)
Short URL: ia.cr/2017/1234
[ Cryptology ePrint archive ]