Secure Computation on Floating Point Numbers

Secure computation undeniably received a lot of attention in the recent years, with the shift toward cloud computing offering a new incentive for secure computation and outsourcing. Surprisingly little attention, however, has been paid to computation with non-integer data types. To narrow this gap, in this work we develop efficient solutions for computation with real numbers in floating point representation, as well as more complex operations such as square root, logarithm, and exponentiation. Our techniques are information-theoretically secure, do not use expensive cryptographic techniques, and can be applied to a variety of settings. Our experimental results also show that the techniques exhibit rather fast performance and in some cases outperform operations on integers.