Advancing Privacy and Accuracy with Federated Learning and Homomorphic
Encryption
Abstract
In this paper, we present an integrated framework that combines
Federated Learning (FL) with Homomorphic Encryption (HE) using the
Artificial Intelligence (AI) models and the Cheon-Kim-Kim-Song (CKKS)
algorithm to address the challenges of privacy and accuracy. FL
facilitates collaborative training of Convolutional Neural Network (CNN)
and Multi-Layer Perceptron (MLP) models across decentralized devices,
allowing for data privacy preservation without sharing raw data. The
integration of the CKKS algorithm for HE ensures secure computation on
encrypted data during the FL process. Our experimental results on three
diverse datasets demonstrate the efficacy of this approach, achieving an
impressive highest average accuracy of 97.3%. Additionally, the CKKS
algorithm is used to achieve efficient computation, making it a
promising solution for privacy-conscious machine learning applications,
and paving the way for practical deployment in various real-world
scenarios, thereby revolutionizing the landscape of privacy-preserving
machine learning.