Graduate School and Research Center in Digital Sciences

SwaNN: Switching among cryptographic tools for privacy-preserving neural network predictions

Tillem, Gamze; Bozdemir, Beyza; Önen, Melek

SECRYPT 2020, 17th International Conference on Security and Cryptography, 8-10 July 2020, (Virtual Conference)

The rise of cloud computing technology led to a paradigm shift in technological services that enabled enterprises to delegate their data analytics tasks to cloud servers which have domain-specific expertise and computational resources for the required analytics. Machine Learning as a Service (MLaaS) is one such service which provides the enterprises to perform machine learning tasks on the cloud. Despite the advantage of eliminating the need for computational resources and domain expertise, sharing sensitive data with the cloud server brings a privacy risk to the enterprises. In this paper, we propose SwaNN, a protocol to privately perform neural network predictions for MLaaS. SwaNN brings together two well-known techniques for secure computation: partially homomorphic encryption and secure two-party computation, and computes neural network predictions by switching between the two methods. The hybrid nature of SwaNN enables to maintain the accuracy of predictions and to optimize the computation time and bandwidth usage. Our experiments show that SwaNN achieves a good balance between computation and communication cost in neural network predictions compared to the state-of-the-art proposals.

Document Bibtex

Title:SwaNN: Switching among cryptographic tools for privacy-preserving neural network predictions
Keywords:privacy, neural networks, secure two-party computation, homomorphic encryption
Type:Conference
Language:English
City:
Date:
Department:Digital Security
Eurecom ref:6259
Copyright: Scitepress
Bibtex: @inproceedings{EURECOM+6259, year = {2020}, title = {{S}wa{NN}: {S}witching among cryptographic tools for privacy-preserving neural network predictions}, author = {{T}illem, {G}amze and {B}ozdemir, {B}eyza and {\"{O}}nen, {M}elek}, booktitle = {{SECRYPT} 2020, 17th {I}nternational {C}onference on {S}ecurity and {C}ryptography, 8-10 {J}uly 2020, ({V}irtual {C}onference)}, address = {}, month = {07}, url = {http://www.eurecom.fr/publication/6259} }
See also: