[Resource Topic] 2024/648: Encrypted KNN Implementation on Distributed Edge Device Network

Welcome to the resource topic for 2024/648

Encrypted KNN Implementation on Distributed Edge Device Network

Authors: B Pradeep Kumar Reddy, Ruchika Meel, Ayantika Chatterjee


Machine learning (ML) as a service has emerged as a rapidly expanding field across various industries like
healthcare, finance, marketing, retail and e-commerce, Industry 4.0, etc where a huge amount of data is gen-
erated. To handle this amount of data, huge computational power is required for which cloud computing used
to be the first choice. However, there are several challenges in cloud computing like limitations of bandwidth,
network connectivity, higher latency, etc. To address these issues, edge computing is prominent nowadays,
where the data from sensor nodes is collected and processed on low-cost edge devices. As simple sensor
nodes are not capable of handling complex computations of ML models, data from sensor nodes need to be
transferred to some nearest edge devices for further processing. If this sensor data is related to some security-
critical application, the privacy of such sensitive data needs to be preserved both during communication from
sensor node to edge device and computation in edge nodes. This increased need to perform edge-based ML
on privacy-preserved data has led to a surge in interest in homomorphic encryption (HE) due to its ability to
perform computations on encrypted form of data. The highest form of HE, Fully Homomorphic Encryption
(FHE), is capable of theoretically handling arbitrary encrypted algorithms but comes with huge computational
overhead. Hence, the implementation of such a complex encrypted ML model on a single edge node is not
very practical in terms of latency requirements. Our paper introduces a low-cost encrypted ML framework on
a distributed edge cluster, where multiple low-cost edge devices (Raspberry Pi boards) are clustered to perform
encrypted distributed K-Nearest Neighbours (KNN) algorithm computations. Our experimental result shows,
KNN prediction on standard Wisconsin breast cancer dataset takes approximately 1.2 hours, implemented on
a cluster of six pi boards, maintaining end-to-end data confidentiality of critical medical data without any re-
quirement of costly cloud-based computation resource support

ePrint: https://eprint.iacr.org/2024/648

See all topics related to this paper.

Feel free to post resources that are related to this paper below.

Example resources include: implementations, explanation materials, talks, slides, links to previous discussions on other websites.

For more information, see the rules for Resource Topics .