Blockchain-Based Federated Learning with Malicious Attacks in Fog Computing Networks

In fog computing networks, the discrete deployment of a large number of fog nodes (FNs) and the frequent information exchange pose significant risks to user privacy and network security. Federated Learning (FL), as an emerging distributed machine learning framework, provides a promising approach to privacy protection. However, non-independent and identically distributed (Non-IID) data and malicious attacks inevitably lead to a decrease in the performance of FL. In this paper, to ensure the security and effectiveness of FL, a Directed Acyclic Graph (DAG) blockchain-based FL (DAGFL) algorithm is proposed. Firstly, to reduce the involvement of malicious models in global aggregation, we propose a device selection algorithm based on reputation and user activity degree, along with an outlier-based malicious model detection algorithm. Besides, to mitigate the negative impact of Non-IID data, a weighted FL aggregation algorithm is introduced, which assigns weights to local models by evaluating their contribution and accuracy. Furthermore, DAG blockchain technology is applied in FL to bolster the security and efficiency of model sharing among FNs. Finally, simulation results show the effectiveness of the proposed algorithms.