Researchers have developed a method to speed up privacy-preserving computations, allowing multiple parties to jointly perform calculations on sensitive data without revealing their individual inputs—and without any third party being able to reconstruct the original data from the results. Details are reported in the International Journal of Applied Cryptography.
The team has focused on a key technical challenge in the field of cryptography. How to perform floating-point arithmetic securely and at scale. Floating-point arithmetic is used by computers to handle real numbers, whether very large or very small, using a codified version of scientific notation so that long strings of zeroes to make a number very small or very large are not needed. This latest advance has the potential to accelerate research and development in privacy-sensitive domains such as health data analysis, financial modelling, and machine learning, without compromising confidentiality.
Secure multi-party computation (MPC) and homomorphic encryption are techniques designed to address the problem. They allow computations to be carried out on encrypted or distributed data, so that no party learns anything about the inputs from the other parties beyond the final result. However, MPC systems struggle with floating-point numbers in terms of performance. They could use fixed-point numbers instead, but that would compromise the precision of the data through rounding errors.
The new research directly addresses the problem by introducing a more efficient way to perform secure floating-point addition, one of the most fundamental operations in numerical computing. The key innovation lies in a protocol that allows many additions to be carried out simultaneously, rather than one at a time. In tests, the approach is 13 times faster than state-of-the-art techniques. Moreover, it uses existing MPC frameworks. The protocol preserves the privacy of all intermediate values and requires no specialised hardware or novel cryptographic assumptions.
Veugen, T., Wezeman, R., Amadori, A., Bootsma, S. and Kamphorst, B. (2025) ‘Secure addition of floating points’, Int. J. Applied Cryptography, Vol. 5, No. 5, pp.1–11.