Secure Federated Learning Framework for Distributed Ai Model Training in Cloud Environments

Abstract

A new design for a secure FL solution that can be used for AI model training across many cloud services. The framework built in this paper is founded on federated learning which only involved the model and not the raw data of the organizations involved. These are federated model training pipelines, methods and approaches of privacy-preserving model updates, and differential privacy and identity and access management. In this framework the issues of leakage of data, privacy violation and invasion by a nasty attacker is also addressed apart from the issues to do with legal matters. Thus, with the help of cloud environment integration and the availability of various data types, the proposed framework allows the improvement of the AI model performance and generation of the models that are closer to real-world while taking into consideration the data sovereignty and privacy concerns.