Swarm Optimization-Based Federated Learning for the Cyber Resilience of Internet of Things Systems Against Adversarial Attacks

Federated Learning (FL) is a paradigm of distributed machine learning that enables multiple devices or clients to work together in training a common model while keeping the privacy of individual data. However, FL has several issues such as slow convergence, communication overhead, and vulnerability to adversarial attacks, particularly in Industry 5 environments such as the Internet of Things (IoT) and its integration with traditional manufacturing processes. These challenges stem from the diverse and non-IID nature of data distributed across clients, which leads to slow convergence and increased communication rounds. This paper aims to address these challenges by proposing a grey wolf optimisation-based federated learning (GWOFL) approach for offering resilience in Industry 5.0 settings against adversarial attacks. The proposed approach decreases the number of communication rounds, reduces the payload between clients and the server, and withstands adversarial attacks simultaneously. It also reduces communication overhead and successfully defends against data poisoning attacks. Experimental results have revealed the efficiency of the proposed approach in overcoming the challenges of FL using the MNIST and CIFAR-10 datasets. The proposed approach converges faster, along with higher accuracy compared with the peer FL methods.