Using Multi-Level Federated and Lightweight Deep Learning Assisted Homomorphic Encryption Based on AI Technology to Provide End-to-End Security and Privacy Preservation for Multi-Cloud Environments

Abstract

As cloud computing (CC) expands, security and privacy become more important. The main problems of increasing processing overheads, communication bottlenecks, and noise budgets are addressed by this study's novel paradigm. Enabling end-to-end security and privacy of cloud-IoT user data and servers in a multi-cloud context is the main goal of this project. This paper discusses the three essential phases of this approach, which improves data privacy and operational efficiency. We start by examining new cloud-IoT users via a three-factor authentication procedure. These users are registered with Trusted Authenticated Servers (TAS), authenticated with their password and user ID, confirmed with photo tags and concealed lines, and have their biometrics examined. Verified users get encryption/decryption keys generated by the Improved Key Generation Process (IKGP). Second, the Trading based Evolutionary Game Theory (TEGT) method is used by the TAS to choose the best DEdS in the Secure MLFL Entities Selection process. We used a machine learning system called Improved Artificial Neural Networks (IANN) to choose our clients. Third, MLFL for Homomorphic Data Sharing and Storage with Secure Privacy Awareness. It employs the Homomorphic Encryption Responsive Lightweight Residual Network with Energy Valley Optimizer (HER-LresNet-EVO) for Local Model Generation. In Global Model Aggregation, Lightweight Factorized Pyramidal Networks (LFPN) are employed. Using MDCS, Super Model Aggregation & Secure Distribution makes aggregated models available to cloud-IoT users. These models are securely kept in the cloud database. Metrics like reduced malicious traffic based on an increase in users, enhanced model accuracy across epochs and rounds, faster encrypting data in relation to data size, optimized homomorphic encryption activities within noise budgets, and a decrease in risks to privacy in comparison to attack ranks are used to gauge the success of the study.