AELGA-FHE: An Augmented Ensemble Learning Based Genetic Algorithm Model for Efficient High Density Fully Homomorphic Encryption

Abstract

Fully homomorphic encryption (FHE) is defined as an encryption scheme that allows arithmetic & logical operations on ciphers, and yields the same effect on decrypted data. Resulting in the widespread use of FHE model for enforcing data privacy in organizations using polynomial arithmetic during encryption and decryption process, and use of different moduli for plain and cipher data. A wide variety of FHE algorithmic implementations are existing and each of these have their own nuances and limitations. It is observed that most of the existing approaches are context-insensitive and do not consider application-specific encryption strength requirements. In order to integrate context-sensitivity and improve encryption strength, These texts offer a Genetic Algorithm design (GA) model for FHE parameter optimization. Results of the proposed GA-FHE model are validated on multiple applications and are stored with respect to their strength classes. These classes, along with their respective FHE configurations are used for training an ensemble deep learning model. This model uses a combination of k Nearest Neighbors (kNN), random forest (RF), linear support vector machine (LSVM), linear regression (LR), and customized 1D Convolutional neural network (CNN) classifiers for strength estimation. Each incoming FHE request is evaluated on these models and their results are augmented to evaluate final FHE moduli parameters. The proposed AELGA-FHE model is tested on a wide variety of textual datasets including ‘Cipher text challenge’, ‘National cipher challenge’, & ‘Secondary cipher challenge’, and strength selection accuracy results are evaluated, strength of encryption, and computational delay. The proposed model outperforms existing FHE methods in terms of these parameters, thereby showcasing its superior deployment capabilities with respect to improved FHE encryption strength and reduced delay needed for high security encryption.