CoInMPro: Confidential Inference and Model Protection Using Secure Multi-Party Computation

Title

CoInMPro: Confidential Inference and Model Protection Using Secure Multi-Party Computation

Creator

Tiwari K.; Bisht K.; George J.P.

Description

In the twenty-first century, machine learning has revolutionized insight generation by using historical data across domains like health care, finance, and pharma. The effectiveness of machine learning solutions depends largely on the collaboration between data owners, model owners, and ML clients, without privacy concerns. The existing privacy-preserving solutions lack efficient and confidential ML inference. This paper addresses this inefficiency by presenting the Confidential Inference and Model Protection, also known as the CoInMPro, to solve the privacy issue faced by model owners and ML clients. The CoInMPro technique is suggested with an aim to boost the privacy of model parameters and client input during ML inference, without affecting the accuracy and by paying a marginal performance cost. Secure multi-party computation (SMPC) techniques were used to calculate inference results confidentially after sharing client input and model parameters privately from different model owners. The technique was implemented in Python language using the open-source SyMPC library to support the SMPC function. The Boston Housing Dataset was used, and the experiments were run on Azure data science VM using Ubuntu OS. The result suggests CoInMPros effectiveness in addressing privacy concerns of model owners and inference clients, with no sizable impact on accuracy and trade-off. A linear impact on performance was noted with an increase of secure nodes in the SMPC cluster. 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

Source

Lecture Notes in Networks and Systems, Vol-462, pp. 1-14.

Date

2022-01-01

Publisher

Springer Science and Business Media Deutschland GmbH

Subject

Confidential inference and model protection; Machine learning (ML); Privacy; Privacy-preserving machine learning (PPML); Secure multi-party computation (SMPC)

Coverage

Tiwari K., CHRIST University, Bengaluru, India; Bisht K., CHRIST University, Bengaluru, India; George J.P., CHRIST University, Bengaluru, India

Rights

Restricted Access

Relation

ISSN: 23673370; ISBN: 978-981192210-7

Format

Online

Language

English

Type

Conference paper

Identifier

https://doi.org/10.1007/978-981-19-2211-4_1

https://www.scopus.com/inward/record.uri?eid=2-s2.0-85135054262&doi=10.1007%2f978-981-19-2211-4_1&partnerID=40&md5=089e4c8b7eacbd6bd3f0aca9a09b5564