LAUSR

Smart devices in an Internet of Things (IoT) generate a massive amount of big data through sensors. The data is used to build intelligent applications through machine learning (ML). To build these applications, the data is collected from devices into data centers for training ML models. Usually, the training of models is performed on central server, but this approach requires the transfer of data from devices to central server. This centralized training approach is not efficient because the users are much less likely to share data to the centralized data centers due to privacy issues and bandwidth limitations. To mitigate these issues, we propose an efficient hybrid secure federated learning approach with the blockchain to securely train the model locally on devices and then to store the model and its parameters into the blockchain for traceability and immutability. A detailed security and performance analysis is presented to show the efficacy of the proposed approach in terms of security, resilience against many security attacks, and cost effectiveness in computation and communication as compared to other existing competing schemes.