Advancing Privacy in Data Mining: Seamless Homomorphic Searches and Precision-Preserving Encryption

In federated learning (FL) systems, the Cheon-Kim-Kim-Song (CKKS) homomorphic encryption scheme is crucial for preserving privacy while enabling computations on encrypted decimal numbers. However, efficient search operations on CKKS encrypted data remain a significant challenge. This paper addresses this gap by introducing a novel search algorithm optimized for CKKS ciphertexts, significantly reducing client-server interactions in decentralized environments. Our approach integrates parallel computing techniques and a balanced binary tree structure to handle complex datasets like CIFAR10 and MNIST efficiently. We also demonstrate the algorithm's applicability to Convolutional Neural Networks (CNNs) for feature selection and privacy-preserving inference. Comprehensive evaluations show our method's scalability and practical efficiency under various network latencies, advancing privacy-preserving data processing in FL applications without compromising computational efficiency or model accuracy.