A Blockchain-Based Model Migration Method for Safe and Long-Term Federated Learning in Internet of Things Systems

A growing number of IoT devices are turning to federated learning (FL) to provide secure data exchange without compromising user privacy. In addition, new research suggests that blockchain technology might bolster FL security. When data in a BFL cluster is scarce, however, current blockchain-based FL (BFL) systems struggle. Gathering a large number of devices to form a BFL cluster is a straightforward answer. However, because to their geographically dispersed locations and the enormous distances between them, these devices have very high levels of connection delay. Due to the frequent contacts in the blockchain consensus, BFL's poor system efficiency would be caused by the high latency. It is possible for attackers to manipulate local models in FL. Therefore, it is possible for manipulated local models to produce an inaccurate global model. So, to provide safe model aggregation, the suggested system uses a blockchain network. By using a blockchain consensus method, nodes in the network may check that the combined model is legitimate, then add it to the distributed database where it will be safe from tampering. Before utilizing the aggregated model, each cluster may get it from the blockchain along with ensure its integrity. In order to assess how well the suggested framework worked, we ran many experiments using various CNN models and datasets.