Privacy-Preserving Generative AI in Healthcare Systems Using Federated Learning Approaches

Abstract

The research paper focuses on the mechanism of introducing Federated Learning alongside privacy-saving strategies in Generative Artificial Intelligence to healthcare applications. The study analyses the privacy protection/model accuracy trade-off by implementing Differential Privacy and Secure Aggregation. The synthetic datasets were applied to the model to train in five rounds and included many federated clients, that improved its accuracy by 49% to 60%. The findings suggest that Federated Learning has the potential to improve the performance of AI and preserve the privacy of data at the same time. Other challenges covered in the study include mode collapse and privacy-utility trade-offs, and recommended solutions to achieve the efficient and secure healthcare AI models.