Privacy-Preserving Collaborative Framework with Auditable Federated Learning

Abstract

This research study provides a privacy and auditable federated learning scheme that guarantees a secure and decentralized machine learning collaboration. The framework can achieve transparency and accountability in federated learning systems through the incorporation of differential privacy, secure aggregation, and blockchain technology. The research tackles the issues of privacy preservation, the model accuracy, and auditing, using the latest privacy methodology such as the different privacy methods and secure multi-party computation. The trade-off between privacy and model performance is experimentally shown to be present and the blockchain offers secure model updates that are auditable. This framework is especially relevant in privacy susceptible industries like healthcare and finance where transparency and data security are the most important of all.