Federated Learning and Privacy Preserving Ai: A Comprehensive Overview

AI has gained currency in recent times, where statistical models seem to be at the forefront. Yet, there are increasing privacy and legal concerns with the conventional centralized training and inference approach to AI technologies. Federated learning (FL) comes onto the set to actually solve such issues revolving around secure distributed AI. It integrates security features at different stages of data including pre processing, training, evaluation, and deployment using secure multi party computation, and differential and hardware privacy. FL aims at addressing issues of data privacy as well as enacting measures to mitigate against data isolation. Nonetheless, federated models are yet to evade threats like plagiarism, illegal copying and misuse. In a bid to do this, FedIPR incorporates the use of watermarks onto FL models for proof of ownership and protection of intellectual property rights (IPR). Many papers that emphasize on security still misappropriate FL to refer to unsecured distributed machine learning. For this reason, this paper reemphasizes the concept of FL and suggests SFL to develop honest, privacy preserving AI systems with secure IP rights cross border guarantees. It also gives a succinct picture of threats as well as attacks and defenses applicable in each SFL phase from a life cycle perspective.