Federated Learning Strategies for Privacy-Preserving Machine Learning Models in Cloud Computing Environments

Abstract

Federated learning presents a promising approach for preparing machine learning models while protecting information protection in cloud computing situations. This investigation examines the adequacy of different unified learning procedures, including Federated Averaging, Secure Multi-Party Computation (SMPC), Differential Security, and Homomorphic Encryption, in keeping up show precision while securing delicate information. Through tests conducted in a reenacted cloud computing environment, exactness, security conservation, and computational overhead were assessed. Results appears that Federated Averaging accomplishes the most noteworthy show exactness (86.5%), followed by SMPC (84.2%), Differential Privacy (83.8%), and Homomorphic Encryption (82.1%). Differential Privacy offers the most elevated level of protection assurance, whereas Unified Averaging shows the least computational overhead. These findings underscore the importance of taking into account such trade-offs between showing performance, security preservation and computational efficiency while choosing unified learning algorithms for privacy sensitive applications.